But that’s just the parts of ML research that take place within your compute environment, namely experiments on model architecture, training strategy, UI/UX, etc. The algorithms and software.

Just because that part of research is progressively accelerated does not imply that we’re on a collision course with world-ending “superintelligence”. It just means that you’ll get to the best possible model of your data faster. And even when your data is the entire corpus of textual knowledge, there is a limit to the utility that provides.

It just means that software engineering, data analysis, and other computational tasks will be less of a bottleneck in the overall progress of the world (which, by the way, is great, and extremely valuable).

I don’t think it’s a stretch to say that we’ll have reached max LLM capability within a couple of years (what the author refers to as “super intelligence”). But it is a huge logical leap to say this:

> If we go through this feedback loop, I expect that the AI systems that come out the other side will be able to very quickly develop novel weapons and very easily overpower human civilization in a conflict.

First of all, the weapons necessary to overpower human civilization exist. They aren’t used as such due to geopolitics and humans’ basic survival instinct. Second, developing novel weapons requires physical tests, which are decidedly not accelerated by LLMs.

AI is a tool wielded by humans. If we’re wiped out by AI, it will be at the direction of a human. But we’ve made it pretty far with some pretty unstable humans in command of planet-killer weapons. I’m not overly concerned.

> The configuration used here had no jailer. Firecracker's production containment includes dropping privileges and restricting the process through namespaces and seccomp. A deployment using those controls requires a separate assessment of the privileges and resources available after compromising its VMM.

If I'm understanding all of this correctly, a properly jailed and configured firecracker instance would greatly reduce the blast radius of this escape, and I feel as though it would have been good for them to explain what jailing and dropping privileges would do to help. I'd hope most production firecracker deployments would be doing this already.

Not saying that this is any less severe and the root issue is not specific to Firecracker, but to a flaw in the kernel's network code -- they just used an insecure firecracker configuration to demonstrate it. However, I feel like some are taking this as a flaw with firecracker when its not. This is an issue in the kernel networking stack.

The Cybertruck I thought was initially kind of neat but when you try to use one as a truck you quickly can see some glaring problems that most people who buy a truck want. Towing with an EV is terrible. I've heard there's power stations with pull throughs but havent' seen them personally. The truck bed isn't practical for anything but a lawn mower. When I ride in one it rides like a car. It doesn't look good dirty. I just think they were going for warthog from Halo and thought the rest would work out. It's also super expensive for all these issues. It's a hobby vehicle. A toy.

I think Tesla is struggling personally. The marketing tactics of showcasing cars we'll never be able to buy just to cause a scene. Don't get me wrong the new Roadster my jaw dropped and I agree with Elon now that SpaceX and Tesla are the 'same' company I think it's great but let's be realistic. Most people want cars that are cheap, dependable and sort of cool. Not everyone can afford multiple cars, and a corvette that isn't feasible. Like I said I work for a competitor (GM) and I fear China is going to eat our lunch. Unless we make infrastructure for charging better and make it a no brainer.

Musk has enthusiasm, yes, but he’s let his luck at PayPal and successfully buying Tesla from the original founders go to his head. Nobody is right all of the time, and no serious project goes perfectly according to the original vision without adjustments. There’s nothing wrong with the idea of an electric truck — companies with better engineering culture are making great ones — but the Cybertruck failed exactly the way everyone who isn’t a superfan expected: it’s designed in a way which guarantees high costs and it just doesn’t deliver corresponding benefits for most people. Diving into the culture wars against a large chunk of potential Tesla buyers was a bad idea but even if he hadn’t, there just aren’t that many people who have that kind of money to spend and the ones who do usually won’t accept so many limitations and flaws.

IME, because of pay, they have a hard time retaining "senior" talent, eg Nvidia, Chainguard, Amutable, Lambda, etc etc etc all have lots and lots of senior+ Canonical engineers. Especially in the age of AI, I think this is a "people-strategy" mistake.

The application process is very long and you absolutely lose great candidates because of it. Salaries are not US-tech competitive, but absolutely are Central/South America and Europe competitive. Lots of great Brazilian and Polish engineers, for example.

Also I have never seen a post on HN, Reddit etc about Canonical/Ubuntu that doesn't complain about hiring process. I think it does a pretty incredible amount of damage to the brand.

3.8 flash and argon have been great FYI. Can't tell a difference between any of the claude models I use at work, minus actually being able to read what it writes.

Antigravity come with Google's models and anthropic - so Gemini 4 Argon, Gemini 3.8 Flash, Claude Opus 5.5 Claude Sonnet 5.5 and GPT-OSS 120B.

Agy is better than Claude code in some ways and vice versa.

Google models are great.

Ironically google music is so terrible I've uninstalled it and use Metrolist exclusively.

I think google models do a decent job helping me put together a blog, for example:

https://hypertekst.net

(Content is typical 90s homepage stuff, not great. But model voice isn't as grating as Claude - and while it has a "look" I think it's s bit better).

Norvig confuses engineering for science. If an engineer wants something that works (and guarantees that it will work) then you use the most effective tool available to achieve that goal. No problem there. But there's absolutely no reason to think that tells you anything scientific about the phenomenon you're modeling. For example, aeronautical engineering says very little about how birds fly. And - currently relevant - LLMs learn to use and understand language, but they use millions of times more training data, and millions of times more computing power, than a child does. So an LLM may be a great technology, but it's probably not a good model for how humans learn and use language.

Chomsky OTOH doesn't see any use for statistics in science, at least not in linguistics. Fortunately for the field of linguistics, Chomsky's influence in linguistics is waning and linguists are beginning to use statistical tools to do science.

At least two of the authors are also avid chiropterologists, with an understood penchant for bunkers and other such spaces.

One summer in my late teens I sat with two of them in an urban park in NW Poland drinking beer, watching bats getting caught in a net half-submerged in water near a bridge and engaging in impromptu comedy with passers-by asking questions like "good catch? any fish?". Good times.

Engaging in field research is a wonderfully rewarding experience, a great way to spend time, meet people and do something meaningful at the same time. If you've never done it - do it. There are plenty of volunteering opportunities - you don't need to be a scientist. Birding, insects, marine life, whatever. Citizen science is also great, but if you get a chance to work alongside scientists, it gets even better. 10/10 good walk, unspoilt.

#55743: Claude Max account email auto-injected into LLM context with no opt-out - May 3, 2026 https://github.com/anthropics/claude-code/issues/55743

#72518: Anthropic embedded spyware in Claude Code - June 30, 2026 https://github.com/anthropics/claude-code/issues/72518

#21151: No indication of WHICH file for READ tool - January 27, 2026 https://github.com/anthropics/claude-code/issues/21151

#42796: Claude Code is unusable for complex engineering tasks with the Feb updates - April 2, 2026 https://github.com/anthropics/claude-code/issues/42796

#73408: AFK auto-continue on unanswered AskUserQuestion is undocumented and has no opt-out - July 2, 2026 https://github.com/anthropics/claude-code/issues/73408

(Also worth a mention, OpenCode PR #18186: anthropic legal requests - March 19, 2026 https://github.com/anomalyco/opencode/pull/18186)

Edit: Oh, and determistic hooks not firing.

Which is great! It's good for the community when people bring first-hand updates here, and to make sure that it continues, it's important that we respond respectfully. (I don't think your post was disrespectful, but commenters have crossed that line in other situations in the past.)

company-helm-research/1.0 (Claude Code agent run by <my name>, <company> platform team; docs lookup)

Claude-Code-research-agent/1.0 (<Company> platform team; on behalf of <my name>; +https://github.com/<company org>/<repo of project>)

Not great, it even pulled the repo of the project I was in.

Your example would look like this:

    export component Hello inherits Rectangle {
        in property <string> text: "World";
        width: 500px; height: 500px;
        background: #505050;
        border-width: 1px; border-color: blue;
        drop-shadow-blur: 15px; drop-shadow-color: #0008;
        VerticalLayout {
            alignment: center; spacing: 12px;
            Text { text: "Hello, \{root.text}!"; font-size: 20px; color: white; horizontal-alignment: center; }
            HorizontalLayout {
                alignment: center; spacing: 8px;
                for c in [#f00, #0f0, #00f, #ff0, #000, #fff]: Rectangle { width: 32px; height: 32px; background: c; }
            }
        }
    }
    

My experience is that it varies greatly by feature.

Something like border styles, the agent was able to do pretty effectively. Although it still took a few hours iterating on it to get the performance characteristics and edge cases right.

More complex features like hoisting position absolute nodes and other layout features still take days to weeks of iterating to get something usable that isn't so fragile it'll blow up the next time you touch something.

I'll also note the frame rate issues in Conker were mainly down to raw RDP fillrate issues, which was for sure the weakest part of the N64 hardware, it could draw beautiful triangles (for the time) but oh so so slowly. There was a decent amount of speech during gameplay using the RSP mp3 codec. The cartridge size was still an issue, 64Mb got a good amount of mono speech but was still very much at a premium.

There's videos from ~15 years ago that I consider gone for good - lost to the great copyright apocalypse - having been unable to find them anywhere on the today's internet.

I suspect that Microsoft has the same history: the entire Windows platform was built with single-user PCs in mind, mostly only dealing with trusted input. The users were also expected to be tech-illiterate for the most part and unable to troubleshoot deep technical issues.

More recently I'm still working on video transmission, but on the production side this time. I also originally adopted the mentality that I should crash early if some unexpected situation was encountered, but I ended up walking that back a lot, because having a critical piece of live media production reboot mid FIFA worldcup final is generally Not Great. Degraded functionality is usually preferable. I still crash early in dev builds, but the production code just logs an error and continues if at all possible.

That said, I definitely agree with you for something like password handling. In this case, I do think it's warranted to crash the program without survivors if something is amiss.

AI suffers the same problem Search Engines had: you cannot ask about something you dont know. Thats why monthly magazines were great: they pushed" knowledge to us so that we learned things we didnt know existed.

[1] https://github.com/binhnguyennus/awesome-scalability

They restricted wage growth too much for too long under Intel (despite being the only growing business unit), so the people with the valuable experience moved elsewhere. The Agilex line-up finally looks to be in the place it should have been 3 years ago, and it seems like they brought in a good marketing team following the Silverlake acquisition, but they still fall in this weird middle-ground now selling "performance-per-watt", which is a hard sell when most engineers optimise for the extremes. If I'm just need a bridge or some glue-logic, I don't need a device which will singe my thumbprints off while doing that. Generally, if you want better performance, AMD does an excellent job, or if you need low power consumption, Lattice is like 4 or 5x lower or more with just enough fmax and IO to be dangerous. If you look read Lattice's shareholder briefings, they're running away with the low/mid-range devices up to ~500k LUTs, and AMD is dominating anything 1M+. I couldn't help but laugh the other day when I saw an Intel reference design with a Lattice MachXO5-NX on it. Look at pretty much any datacentre rack now, and you'll spot 4-64 Lattice devices, when it used to be MAX10s top to bottom. Everything else is swept up by AMD.

For Altera, Agilex 9 (ADC and DAC converters integrated with an SoC FPGA) is probably their most interesting device, but that customer base is limited - basically comms and maybe radar with some weird and wonderful optical or quantum compute applications. I've heard friends of mine in the industry say very good things about it, but RF is not my strength so I can't comment.

In all honesty, they've got an insane fight on their hands to regain the market share. They canned their low end road-map when Sandra Riviera came in (as a classic Intel VP would do) and put all their eggs in the Agilex 7 basket for a long time. They've got Agilex 5 now (and Agilex 3, which is just the bottom SKUs of 5 with it painted over by marketing to fill the gap), which is great for vision applications, but only because they had a fantastic video and vision IP team who last I heard have left for ARM.

I hope they pull it back, as I loved the people there and it had a great culture, but it's been awfully mismanaged for a while and I question how easy it will be to pull it back considering the monopolies and duopolies that are appearing in the semi industry right now.

I think the argument is pretty simple:

- Plastic is bad (environment, microplastics, etc) - We should minimize it

That doesn't necessitate a "all the way to zero or don't bother" logic, which what you're saying reads as.

Coffee pods machines have been an environmental nightmare, great to see a growing trend that addresses this.

Central planning is the status quo. That's essentially what zoning rules are, and why everything is such a mess. You want to open a shop and live above it? No, sorry, not zoned for residential; banned. Operate the shop out of your house instead? Haha, banned. What if your shop really doesn't need very much space at all, so something the size of a closet would do as long as it's in the right location? Nope, minimum unit sizes or similar, prohibited by law. And so on.

If they want to stop doing this that would be great, but if they're not going to stop doing it then having the rules be things like "no chains" instead of "no small units" would certainly be an improvement.

Our neighbor's mom called the animal warden on us for not feeding him. He eats better than I do.

Her dog is a morbidly obese miniature collie.

Most people's perspective of ideal animal shapes is incorrect.

It is quite possible that flies prefer one part of the light spectrum more than another part, but to say in the title that they don’t swarm at LEDs is misleading at best.

But as is, it's probably hard to become popular amongst hobbyists and local model enthusiasts like Qwen's 27B class models or MOE models because it's a 65B dense model and inference will be prohibitively slow on consumer grade hardware, and seems like the goal/mission of the company behind this model is to automate chip design.

So if I have to assume and connect the dots, this is for the companies who want to use AI in air-gapped environments for chip design so that they protect their IPs (not leaking to third party AI labs/hostile parties) and can afford lots of server grade inference hardware for a quick unlock with the big dense model.

Neat!

I like that they acknowledge that. For working with ai as a tool open models are already great. For millennium problems and doomermaxxing they are not there yet.

The rule includes >90% of "remote" workers who have discovered they were classified as hybrid (a classification now eliminated) and their manager didn't make it official. Now, remote approvals are SVP level.

The issue is, AMD still pays significantly less than any company with an RTO mandate, even with the stock appreciation. And the stock has gone down 25% in the past few weeks so it's even less competitive. The typical engineer was not given an equity-weighted package, even L9 only 30% of your TC is in stock.

It's a recruiting bonanza right now and everyone's getting calls offering 50% pay bumps from 5-day office semi corps like Broadcom or Qualcomm. The better people are getting offers at Google or Apple for 3-day RTO for even greater TC bumps.

In the last few years Intel's division that makes Xeons had huge losses, because in order to retain their big customers they had to accept to sell at big discounts, otherwise the customers would have switched to AMD.

All has changed this year, due to the AI craze.

AMD has sold all the server CPUs that it could make, within its allocations at TSMC. Everybody who did not succeed to buy EPYCs has bought Intel Granite Rapids Xeons, because they are not made at TSMC, but in Intel's fabs, so they did not compete with other TSMC customers for allocations.

So Intel has also sold every server CPU that they could make, and they are now trying to retool some fab that made new Panther Lake CPUs back to the obsolete Intel 3 CMOS manufacturing process, in order to make more Xeons, to sell them as long as this huge demand persists.

Thus in the last quarter both AMD and Intel have sold less consumer CPUs than in the past, because all who could did delay their upgrade plans due to the unacceptable memory prices, but on the other hand both AMD and Intel had record revenues and profits from selling server CPUs.

At some point students started painting their own crosswalk, which would be removed again and again by authorities, until the city eventually got the hint and installed a permanent crosswalk with a traffic light and a pedestrian island.

Sometimes the way to get something is fixed is by demonstrating the fix.

- Any new datacenters being built cannot raise utility rates

- Some of the revenue generated by the datacenter must be set aside to improve power reliability

- Datacenters must not touch our waterways at all

- Datacenters must be built using union labor”

Deal. This sounds like a great list. I would add, “no property tax breaks”.

Just get the state to pass this as a law and set appropriate enforcement mechanisms and then just let data centers go ham if they meet this criteria?

No more “per community meetings” and municipal haggling (see: corrupting) local governments. Buy the land build the data centers. The municipality can figure out the exact benefits that works for that community from the giant tax coffers.

This seems imminently fair to me.

* We've built a technology that has obliterated entry level hiring

* That technology might obliterate all the other "normal" person white collar jobs. The kinds of jobs you can clock in and out and not think about them outside of that. Instead, every new job will be a managerial one working with 500 agents.

* Loaning all the money to build these data centers and train these models costs so much it cranks rates through the roof, so we've frozen the real estate market and nobody can move or buy in

* The past 10 years of American politics have been defined by how the last time we did a broad replacement of labor event like that for a group of people. Their standard of living permanently shrunk (the NAFTA disrupted working class), and we may end up doing that to you guys again. The only solution is that the government is going to have to do some sort of UBI or redistribution (which the last time they did during COVID, it was followed by the greatest inflation we've seen in our lifetimes)

* This technology might literally destroy the world in the next 10 years, and we keep accidentally releasing it into the world like its Skynet. But we are not responsible for it, and by the way its unavoidable with this technology. Somewhere on the other side, utopia will arise (No really your standard of living will actually go up after losing your job this time unlike everyone else, trust me.)

Then half the people in this thread is smugly are acting baffled at why people don't want this, and going "Hey they don't know that much about data center logistics but still oppose this. What is up with that?." Forget about if a data center actually ends up being a net drag if its near you and the standard nimby stuff. Even if it weren't, what benefit does marching towards this future have for anybody not in the top 10%? Sure, you can make a bunch of economic theory arguments: growth is always good, stagflation, Luddites, etc. But in practice, economic advancement and technology over the past 30 years has advanced so fast people can't even comprehend it, and in return everybody is becoming stupider, more unhappy, polarized, and the median standard of living has dropped as rich people have 20x their net worth. Why would a normal person want more of this?

* https://en.wikipedia.org/wiki/Kepler%27s_laws_of_planetary_m...

People did not understand how or why they worked, but they switched over to them from the Ptolemaic (Tychonic, etc†) system anyway because the math was easier. It was only later (Newton, Einstein) that we understood the "why" and "how".

† https://tofspot.blogspot.com/2013/08/the-great-ptolemaic-sma...

I don’t think anyone “allowed” them - quite the opposite. They’ve made generational bets that ended up being right, bets that IIRC the industry did not agree with for a long time. Them pursuing CUDA resulted in a draw down of something crazy like 75% of their valuation. They bet the farm multiple times to get to the point they are - and it’s only “obvious” in hindsight.

If you want folks to make generational bets as a matter of course (which I think most folks do), centralizing the downside of those bets and socializing the upside of those bets is not a great strategy. Almost as bad as socializing the downside.

Competition will eventually close that gap. We’re only a few years into this thing.

Nothing stops you from giving Nvidia half a million for their latest and greatest AI GPUs, unless you are in China I guess, but China does very well despite the export control.

It is just classic market segmentation, the thing Nvidia and just about every company that can benefit from it has done for decades. Cloud providers are ready to pay a lot of money, consumers wanting local AI have a lower budget, and gamers even lower. But a low budget is greater than zero, and there is still money to be made on these markets. But they also don't want to compete against themselves, so they are going to find ways of crippling enterprise-class workloads on consumer hardware.

It used to be that they crippled high precision arithmetic on consumer hardware, because it was a differentiating factor, but AI doesn't care about high precision, so now, it is something else.

Sovereign AI has never been more important.

Controlling AI is so much more powerful and impactful than controlling search or social algorithms.

Centralized AI a 10x on social control over everything we have seen before, and there is no way that a free and open society survives contact with centralised, high capability AI if the system is manipulated towards guiding outcomes. It’s essentially mind control, as people will offload an even greater share of their cognition to AI than they do with search and social.

Build and maintain sovereign AI while you still can.

https://jcd.pub/2026/11/04/screenshake-shaders-for-kitty/

Both Calibre and Gimp are great examples of outsider design.

Works amazingly well.

Looks amazingly bad.

> Windows leadership should be fired and replaced by people who are actually Windows users, not MacOS users.

As a longtime Mac user, Windows leadership decisions don’t make any sense to me either. I have precisely zero interest in making 11 my primary OS… XP, 7, or 10 wouldn’t have been ideal but if I had no choice but to work with Windows those would be greatly preferable to 11.

One of the best features to be added to C++ in many years.

C++ contracts are an attempt to bring C asserts into the C++ age. The weakness of C++ contracts is that they are mostly syntactic sugar over C asserts. If you really care about this type of thing, you probably already have a custom C++ assertion framework that is at least as capable if not more.

The promise of C++ contracts is that they can aid formal verification, which would be great. As delivered, it can’t do much more than was already possible, which will inhibit adoption.

"Oh, theres a process."

Great. Where is it documented?

"in a spreadsheet."

Which spreadsheet?

"Well, theres the old spreadsheet, new spreadsheet, and the spreadsheet that Karen maintains."

Where's Karen?

"She left in 2021."

Qwen3.8-Flash_Next should be in the running for that title as well. It's astonishingly good. I use both of them. If I need more memory on the machine for other things I run 27B, If I can get by with much less RAM I run QFN (takes about 102GB on my Strix Halo box with 262,144 context). QFN runs quite a bit faster on my hardware - about 50t/s. 27B with MTP starts out around 22t/s but degrades over time as the context fills up. Both are great models, but I'd give QFN a bit of an edge (at least for the coding projects I've worked on). But both feel on par with Claude Opus 4.6 - 4.8.

That said, I'm sure there are architects today for which some or all of this is not true.

Most of the above are subjects an architect student would have had to learn about and demonstrate profienciency in, to earn their degree, a 100 years ago.

I'm sure the change in architecture's priorities played a role in this too. Industrial materials, construction efficiency. The need to mass produce housing, costs and new social ideals, yadayadayada.

And ofc the modernist rejection of any ornament; even that which is there historically, on existing architecture.

My hope is that in a future where we 3D-print buildings or construct them with robots we will see a renaissance of non-functional[1] beauty/sculpture/texture/color in architecture because the cost difference will be marginal/nonexistend.

But then, architecture is also always a mirror of the period. And from that pov it's not looking so great for beauty making it back any time soon.

[1] Yet functional because how can the happiness/joy of the perceiver of architecture be removed from what should count as functional, in architecture?

I think this is underselling it a bit. If you'd told this to the IBM engineers at the time I think they would have taken great interest in how the performance was achieved.

On the same theme:

> the supreme goal of all theory is to make the irreducible basic elements as simple and as few as possible without having to surrender the adequate representation of a single datum of experience

- Einstein

> I would have written a shorter letter, but did not have the time

- Pascal

> Simplicity is a great virtue but requires hard work to achieve and education to appreciate, and, to make matters worse, complexity sells better.

- Dijkstra.

This has become common now, and I've seen it ruin large projects until nobody understands the codebase full of verbose nonsensical code and comments. It becomes impossible to further develop the project without AI. Contributors not using AI for any reason are all leaving because no one enjoys working with mindless robots (and humans) as colleagues. It's demeaning and devaluing their labor and humanity. They're not getting "left behind", they are leaving the madness of crowds behind.

> ..the possibility of runaway dynamics: once a critical threshold is crossed, small increases in adoption can trigger rapid population-level shifts toward *persistent dependence, with abrupt losses in cognitive competence*.

We can all see it happening. As this technology permeates society, the challenge to maintain control, coherence, and quality will become greater and ever more serious - for example, AI in healthcare, military, management and governance.

The challenge took me a few weeks to make because "calibrating" a disk (determining the geometry of the pregroove) is incredibly sensitive.

After the CTF I tried really hard to make a repeatable calibration process with the intent of building something releasable. I even ended up measuring the details on every CD-RW and CD-R I could buy on Amazon.

But despite trying very hard, I never did make calibration patterns that were reliable enough that I felt like other people would be willing to go through the process.

It seems someone else has succeeded where I failed. Great to see!

In general my personal research thesis is that current deep learning is much less malleable and neuroplastic as it should be, I got into interpretability because of this. Many focus on safety, but I prioritize it as a toolbox that allows for greater shaping of the underlying intelligence system. If we can fully understand how learning happens, where information is stored and how it is processed, unlike in biological brains, we can apply edits and transformations in digital ones. And if I can use a technique to edit weights based on some underlying reasoning about a system (with a caveat that it leads to something worthwhile), so can an agent. Hence self-improvement. There are many interesting implication to weight space research, and I very much encourage people to read about it. But it is also very easy to overstate what is currently possible, in fact I'd say better to know what is not so one can make it possible.

Part of my research is in weight space learning. For starters, we have had the ability to edit weights, to generate new weights, to identify weights for a while. Here is an example for larger models as early as 2023: https://rome.baulab.info/

Weight space interpretability and learning is an active research community, has been for the past 3 years with many great papers. I suspect it is hitting non-interp folks harder because the authors hyped up their work, and people who are not in the field did not bother to read it before typing up X posts.

Some experiments here are done on very small models (20K params). We already know that small models are more interpretable. In fact, much of editing here you can do by hand. For Qwen, the model fixed a looping bug that may have been known apriori, and it is basically a steering vector, a very established technique. This is clear when they then attempt to "improve" math capability, and it doesn't work. For me, this paper rather shows that OpenAI has done training on mechinterp environment, in fact, maybe even on weight space learning datasets, which are open. But it does not show some insane "well past what we’d expect from humans or conventional algorithms" (from the blog post), more like past what I would expect from a large model with no specialized harness. I still think it is cool work, but do check out recent advances in autointerp, to ground your expectation.

On a personal note, I understand the need for promotion of your work, but I think healthy skepticism is fine too. Otherwise, your understanding of what problems need to be worked on and solved in science becomes warped.

Additionally, some ETFs have built in protections to rule out businesses that are not yet profitable (like SpaceX and Anthropic), which I think is a great rule.

Physics aside, Chiral Altermagnon is a great band name.

They remain one-stop shop for games. The gamers don't have to weigh down their systems with a bunch of different bloated game stores, each "needing" several GB of memory occupied at all times and adding minutes to machine startup.

Not to mention conflicts between a bunch of different anti-cheat and anti-piracy technologies on the same machine. And stores with bad UX (e.g. that make you suffer additional wait screens or naggings; Steam is as unobstructive as it gets).

Compare that to Netflix: at the begining, it was a "one stop shop" for movies, it had a lot of classic movies and TV for reasonable price. When the concept was proven, the production houses revoked the license for content. Competition came to the sector and each movie production house set up their own subscription based streaming service. You need to keep paying a whole bunch of services if you want to be able to keep up with all popular TV production or watch classic movies.

So, the competition would in theory be great. But even if stores didn't have subscriptions, the gamers don't want to fill their personal data, check privacy settings, configure, troubleshoot, debloat, ... a different app store for each game. When people have a slice of time to play, they don't want to deal with all of these.

I'm sure that when Valve forgets what keeps them successful and start becoming a Netflix, the Steam store will get replaced.

That’s a bingo. Most people meet the Russians who escaped during the post-Soviet brain drain, they don’t meet the people who stayed behind. Many of whom became gangsters and turned the country onto anarchy. They were products of the Soviet education system as much as anyone.

At the same time before we go idealizing that system, what Western folks often see are the best of the best examples. Individuals who went to great schools, emigrated and excelled in the West. They may have gone to better schools, with better teachers, or had parents who cared more etc. They don't see a representative sample. Many schools, especially in smaller cities and villages, were not as great, teachers not as caring, cheating was rampant, bribing teachers even more rampant etc.

They switched from ClickHouse to DuckDB five months ago, then switched from DuckDB to their own Trino derivative?

The article then ends before saying if the new setup solved their problems.

It’s great that we can now make these migrations with little downtime (I assume?) and try different things, but I wish we could have read more about what testing they did before jumping. The fact that they jumped to DuckDB without realizing it wasn’t going to work at their scale and spent only a few months with it before jumping again is not convincing.

I hoe there’s more to this story that just didn’t make it into the blog posts. As written it’s giving me flashbacks to some teams I worked with who changed tech stacks every six months because they liked migrating more than tuning and maintaining.

That ArchiveWeb extension looks excellent for custom captures. Picking the pages and elements you want during capture is a great feature.

For quicker one-offs, SingleFile gives you a single HTML with all the media of a page: https://www.getsinglefile.com

If you're making archiving a more involved part of your workflow, I highly recommend ArchiveBox. It can save pretty much every type of file from everywhere, in almost every format you can imagine, while building a searchable index of the archives. Been using it successfully for many years: https://archivebox.io

Maybe they tried too hard to get good Q2 numbers, making revenue happen earlier and expenses later, and that now makes Q3 look worse? Either way, this does not look great

On our own 1.5M loc system, we've found and fixed thousands of bugs, security holes and circular dependencies in the last year. Code quality has increased, test coverage increased. We've also recently set our compiler to treat warnings as errors, and to be as strict as possible. It is absolutely amazing to see the gaps being filled, it would've been an impossible task for human devs and IDE's would've made it very uncomfortable. We've also ironed out all the bugs and failure modes in our infrastructure automation / monitoring. Our next project is to eliminate all 3rd party dependencies, and after that to rebuild our frontends completely using zero frameworks, not touching npm ever again after that.

So when I read this about postgres, I was very happy that they found and pursued this side quest. They can make the codebase bullet proof.

My point was, you still didn't build infrastructure out of great lumps of clay. You don't slap down some clay in the middle of a stream and vibe-build a bridge. Infrastructure was still built of stone. Or, if clay, out of individual bricks.

Similarly, LLM-generated code should be used to build discrete bricks of code that can be read, tested, and most importantly, understood. Then you take those bricks and architect and engineer your application with them. This is how you build a platform that can survive actual use.

You can make a model of a bridge with the clay...but the actual bridge is an entirely different pursuit. This is a discipline we need to develop as engineers, since LLM-generated code won't go away. It can be used to make bricks faster but we have to make the engineering decisions that create reliable, secure applications.

The world is a very different place than it was in the early 20th century, but at the same time, "those who do not learn from history are doomed to repeat it".

I say this not to be a "doomer". I say this because those forces I mentioned above were very real and we have to keep our eyes wide open.

[1] https://en.wikipedia.org/wiki/Great_Depression

In all of my decades I recall only one telephone outage one time, at home. A truck (or something) had snagged a line feeding the house. It was repaired less than an hour after reporting it.

And I'm nostalgic for that kind of reliability, but it's gone now.

I was chatting with a tech who was installing a new county 911 system a few weeks ago. There was a time, not so long ago, when 911 had been robust: Fixed circuits to multiple different areas (to walk around network issues). Slow, redundant, robust analog modems on leased lines for caller data and call control. Eventually, things like PRI or T1.

Either way, there were great racks of redundant gear at the 911 PSAP for managing calls.

But now? It's just VOIP over the public Internet. 911 calls get the same not-special class of service that posting this comment does. The huge gear-filled racks are gone. In their place is a router and a switch, and a collection of dispatch PCs.

(There's a backup IP connection, of course. This backup connection uses cellular data.)

I used Zocdoc in the past to find a doctor and ended up with a doctor who just charged out-of-pocket instead of insurance, and it worked great. Would an app that shows up-front pricing be better for the consumer?

Code: https://github.com/fyquah/ramear

Press: https://technical.ly/software-development/scary-hardware-hac...

The most challenging part was to demonstrate it live on stage, typing on the airgapped desktop and decoding the signals on the laptop with the antena next to it. I found an image still surviving from the stage: https://dailytrojan.com/2016/01/26/my-weekend-at-pennapps/ I may have a video of the demo on the stage as well...

Our work was based on the same author Mordechai Guri's earlier work 5 months prior, referenced on GitHub (USENIX Security 15, pp. 849-864). Thank you sir, this hack was a highlight of my student years.

OTOH, it's chilling me to this day. The implications of this attack vector is only getting scarier. Also, apparently the capability ceiling: 1Kb/s and a greater distance is demonstrated here compared to 11 years ago.

WASM is a non-starter for most web applications due to initial binary download.

Advising to use some experimental feature in one browser does not seem like a great idea.

JavaScript frameworks are not perfect but they exist for good reasons. People that brush them away usually lack the experience and understanding on the topic.

Five years ago I was spending up to €1000 a day in Google ads for my star-map company, Greaterskies. After a few months of meetings with an ever-rotating group of "experts" working for Google, and advising me in my ads strategy, I had reached a point where most of my revenue was going to Google. What was left was barely enough to keep going.

This is what capture feels like. You still own your company, but you are essentially an operator making money for someone else.

I think this will be the fate of people who allow their processes to be captured by the model providers. Always run at least two harnesses, and make sure that they are swappable.

If I didn’t have a subscription that doesn’t cost me anything, I would use gpt instead. At least gpt models don’t lecture me constantly.

- Refusal to log me in to websites in remote browsers. I was in the mountains and urgently had to submit something so I wanted to remotely use Claude on my Mac at home to help me but it told me it will never log me in because it won’t impersonate users

- Refusal to repeat XHR requests automated in a browser for me to do something (“no, I will only do what a user would do I’m only gonna click around”)

- refusal to enter ANY credential into any scripts, even dummy or dev tokens (“you have to do that yourself. I won’t do it and that’s the last I say of this”)

- constant [cyber] for very small tasks when it becomes a tiny bit technical, even though I’m in the cyber program. With gpt or glm/deepseek I can at least ask cs questions or explore binaries

And the list goes on. Every day it’s a coin flip whether it complies with something or won’t and I have to use OpenCode Go instead that I have as backup now.

At this point Claude is good for coding and that’s it. It’s a tool that doesn’t do what I tell it to and gives me snappy remarks from his opinionated morale high ground. I went from full on Claude maxi to hating everything they did with the 5.5 generation

Absolutely. I have been [cyber]'d so many times even Claude jokes about it now. Claude is great but it's operators are making it unreliable for me. I have moved one of my projects elsewhere and will move the remainder tomorrow as a stop gap until I get the right hardware set up to run abliterated local models for some percentage of my needs. Their message has been received loud and clear.

Now this project gives me some ideas of things to implement:

- pick the thumbnail from the middle of the video

- allow curated subscriptions, i.e. I subscribe to some curated list made by someone (in this case it could be the database of videos Karpathy has gathered from HN, but there could be others)

At some point I was almost implementing DeArrow support, but using it for a while I don't think it is too great, often the alternative titles people give are unreliable, and honestly I don't know how they exist without being spammed or trolled to death. I do have SponsorBlock support though (I feel like SponsorBlock would work better if it allowed high-quality AI bots to submit blocks though).

Based on the name I think ur (ancient) Dinkel (spelt) is even older than regular spelt but I'm not sure. All I know is that it makes great crackers but terrible pasta and ok bread but not great.

The CIA sponsored extensive studies on this and came to the conclusion that democracies with a strong presidential role are associated with a greater risk of failure. It's not really surprising, and it can be seen in the US right now.

The problem is, the moment you use `async`, your state machine reduces down to one API: `poll`. What if I want to change some state in-between two state transitions (i.e. between two `await` points)? Now I have to wire-in a channel to supply that data at the right point or access some shared data via a Mutex.

Rust's `async` implementation is great if you know everything upfront before you kick off your state machine. The moment you need to do many things concurrently, it can become a real mess.

Firezone's sans-IO core handles many things concurrently: inbound IP, inbound UDP, WebSocket control messages, TURN allocations, timers, user commands, etc. Having essentially one big struct that I can just call `&mut self` functions on for any one of these inputs makes this very easy to understand, compose and test.

A few things like TURN request-response handshakes might be easier as an `async` function but in reality, it is not a big deal and the composability you get from it outweighs it in my opinion.

> The design also avoids problems like function coloring and plays well with Rust's ownership rules.

Async is not I/O. It can be used for I/O, but it's just a way of writing pure, deterministic state machines that keeps a logical flow together instead of scattered over a bunch of transitions.

Even if you choose to manually compile your async flows into explicit state machines, you still have to deal with ‘function colouring’ (a.k.a. typing): a flow that crosses several state transitions (and may have other disjoint parts of the state updated in the meantime, via other transitions) now has to be rewritten as a (series of) state transitions, and can't be called from a blocking function. If anything, rewriting all your state machines without `async` makes the function colouring ‘problem’ worse, because multi-transition flows are even more different from normal functions.

Think about it. Google's latest Gemini AI is just behind Anthropic's best offering -- or, in fact, it may be slightly superior. Google also has dozens of other mature, cash-cow businesses, massive infrastructure, their own hardware accelerators, an army of smart technical staff... and leadership that, for all its faults, isn't a literal cult.

Anthropic has a top-tier LLM among other top-tier LLMs. Claude is great, but not unique. Anthropic has effectively no other assets, many genuine liabilities, a downright bizarre wet-lab venture, and leadership that's... well...

So, the way I see it, either Google is massively undervalued, or Anthropic's IPO is massively overvalued. Am I somehow off base here?

What works: - Empowering engineers. Giving engineers more trust, ownership and product decision-making leads to more engaged engineers vs. micromanaged tickets. Engineers who are treated like owners tend to notice cross-cutting problems and push to get them cleaned up. - Hire for diversity. Not race or gender, but diversity of thought. Different ages, backgrounds, approaches to AI, focus on frontend/backend, etc. Race/gender diversity tends to come along for free. This is a huge edge because every LLM is spinning up the same next token prediction, so approaching problems from different perspectives gives more robust coverage across your codebase. - Flat org structure, low ego. Interns and 20 year vets work side-by-side and teach each other things - hire for openmindedness. - Communicate about product. What should it do? How do we want a feature to work, and how to handle certain edge cases? If the product is well understood, the code follows. In-person/hybrid is valuable. - Architecture matters more than ever, but it is easier than ever to fix. Listen when engineers mention tech debt. Notice when a simple change touches 27 files. Senior engineers shine for this, because they tend to have a better nose for code smell. - Reviewing still matters. Anthropic may throw tokens at reviewing, but it is a great way to share ambient product/code knowledge across the team. 200 engineers throwing code against the wall is chaos and leads to YOLO codebases. You can move fast AND keep a rough mental model of the codebase intact.

And it's devouring tens of millions. I know what tens of millions can make in the video game industry, it can keep a small team of good devs and designers and artists making something special that's _really_ enjoyed by tens of thousands of people.

We don't want more games, we want more great games. This will never, ever, make great games.

https://www.spawn.co

Look on ye mighty, and despair. I really don't care how this sausage was made because the only thing you need to know is that it shouldn't ever have been made.

I'm sure those writers were quite content writing for an actor of that calibre. Imagine putting words to the page, knowing Patrick Stewart will take them out into the world and surely stick the landing.

What a great combination.

We know Stewart was not a great writer, of course. When he had opportunities to take the reigns of the Picard character creativitely he usually asked for more action, and the results were a bit more conventional and boring. The combination was certainly also needed for the magic.

I always liked this video from an acting workshop where Suchet and Stewart deliver contrasting performances of Shylock and discuss their choices at length:

https://youtu.be/-35G7DOFrOo

https://youtu.be/FU_zqBIITDM

If AI can handle it, great! We have a new tool -- let's make the tool better.

Let's figure out how to use AI to make us detect and resolve issues faster, and even better, prevent issues. I've always said we can't consistently hit 5 9's without automatic detection and remediation. We are there now.

Let's get everyone to 5 9s with AI!

This is a great example. Aviation is a perfect industry to learn from, they figured out a lot of things about how to keep humans from making mistakes. While complexity increases and automation takes away more and more tasks.

So yeah, really not exciting ... just sad.

First off, Opus 5.5 built working weights and code with only two prompts:

1. Read this article and any relevant source code: https://divagr.com/blog/qwenputer. Then explore if you could emulate the chip and ram in a Nintendo Entertainment System (as people do today with FPGAs in MISTRs) using this technique. If so, what size model would you have to build and if you could use smaller faster ones, could you get it to be at the standard clockspeed.

It returned charts well illustrating any model would be _slow_. But it's still fun, so the second prompt was:

2. I think we should build it! Just for fun. It's okay if it's not fast. We want to be able to have the weights, and a cli to run them that we can point at a ROM.

Now while this was an _epic_ run by Opus (5.5 is a crazy capable model), this is a surprisingly well-structured challenge.

There is a long-lived and incredible community of classic gaming emulation work online for it to reference. And that community has created…conformance tests! Opus found them and built against one, that has >3k golden instructions: https://github.com/christopherpow/nes-test-roms/blob/master/...

Also, there is the original Qwenputer repo that Opus used to drive the weight writing. That was essential.

But this is a great example of what these models can do with proper reference materials, great specs, conformance tests, and clear direction.

Then day | light break | …… [And] WILD | MAN make | of he | mouth a | shape… he say | to DIC | TATOR ‘Listen | to the | dream I | dream in | the night The sky | god | + + + the | war god | the god | of know | ledge and | the sun | god all | meet to | gether | in the | great court | of the | sky… And the | sky god | AN + + + | say to | the war | god… “Because | they kill | the MAN | COW of | the sky | + + + and | because | they kill | TREEGUARD the one* | who cut | down the | hard wood | tree on | the moun | tain side | must die”

It shouldn't be your first read of Gilgamesh, but its a fun, insightful second.

- much more awkward to use

- more liable to hacks

- deposits are not insured

- fewer legal protections

- doesn’t pay interest (or not easily anyway)

- and as I find out, bannable by a central institution with lower legal oversight than the Fed

I’ll keep a mental note of this.

[0]: https://bsky.app/profile/molly.wiki/post/3mz27m5slcc2o

This just seems to me like (1) not wanting to learn new things, (2) a logical fallacy because the conclusion that those changes are to "stay relevant" and not to make the life of any future mantainer easier, and (3) fear of the new (https://en.wikipedia.org/wiki/Neophobia)

Why? Why? Why? I hate how everyone seems convinced they need to adopt the latest and greatest browser features just to stay relevant. I hate oklch and I hate Tailwind popularized it. I don't care how scientifically accurate it is at representing human color perception. It's a PITA to understand a new type of color system, and learn to tweak it, especially when HSL/RGB just worked fine.

https://www.nikonsmallworld.com/galleries/2026-photomicrogra...

In Zig app code you don't have to pass allocators all over if you don't want to, that's your code so do what you want. But for libraries it's the right thing to do because it puts the caller in control.

The reason you have to pass allocators instead of making this implicit like in e.g. Odin, is because Zig just takes a firm stance against implicit logic, firmer than any other language I've ever used. The end result is that people who like Zig tend to like passing allocators around even in their own code.

God what an annoying thing to say. Suggesting that movies will essentially become games, or that future video games will be unimaginably better (presumably because they are driven by generative world models) betrays such a surface level appreciation for what makes art and culture and all this humanity so beautiful as it is.

The entire point of movies is the deliberate choices of the filmmakers! Movies will not be improved by our ability to choose what the main character looks like or add a style filter. Super Mario 64 is great because it is an expression of the very hardware that constrains it. Inevitable future versions where anything can happen because someone hooked the ROM up to a generative world model won't be better for it.

I agree that this technology is absolutely world changing, and am cautiously eager to see what comes of it, but it just bums me out, perhaps needlessly, that the people at the helm can be so artless, unreflective. Whatever.

[1]: https://sw.kovidgoyal.net/kitty/kittens/hints/

[2]: https://github.com/ghostty-org/ghostty/discussions

[3]: https://github.com/ghostty-org/ghostty/discussions/2394

As a reminder, libghostty is the highly portable, embeddable library that provides core terminal emulation, render state, and more. I always felt that a free, high qualty, lightweight, reusable terminal emulation core would benefit the developer world more than any single GUI application. And that is playing out!

It is and always was my true purpose with the Ghostty project (I said this in my first ever talk about Ghostty!). I've jokingly said many times that Ghostty GUI is just a glorified tech demo for libghostty. That's not fair though, because GUI is great and there are really talented people working on it.

Since I started the project, Ghostty has always gotten people saying "I don't understand why you'd start a terminal emulator today." Well, the answer is libghostty. It's truly unique and its enabling something special (imo!).

Highlights of libghostty since 6 months ago (none of the below was true 6 months ago):

- Libghostty has nearly 100% coverage of all of Ghostty's core (and you can compile with flags to remove any of these features for size/safety): https://libghostty.tip.ghostty.org/

- Libghostty now builds and is tested (in CI) for: Apple (macOS, iOS, visionOS), Windows (MSVC and GNU), Linux, FreeBSD, Android, WebAssembly, and freestanding (i.e. direct on hardware with no OS).

- Libghostty has bindings in pretty much every popular programming language: C, Zig, Go, Swift, Python, Rust, .NET, etc.

- Libghostty's memory requirements have dramatically shrunk to enable the portability above. There's still work to do but a single empty terminal now takes 90% less memory. And at scale I've proven 10,000 active (input/output happening) terminals on a Linux server with 4 cores and 8 GB of RAM.

- Libghostty powers all JetBrains IDE terminals, over 12.5 million active users. https://blog.jetbrains.com/platform/2026/10/jetbrains-termin...

- Libghostty is the most popular core behind the wave of venture-backed "agentic" terminals and editors, such as Herdr, cmux, and many others.

- Libghostty is being discussed and has been verbally accepted to replace the terminal core in Neovim: https://github.com/neovim/neovim/pull/39773 (work still to be done)

- Libghostty powers a really popular and powerful terminal emulator for emacs (Ghostel): https://github.com/dakra/ghostel

- Libghostty's speed is used by Antithesis to fuzz popular terminal programs and has directly led to hundreds of bug fixes sent upstream to programs such as tmux, btop, shells, and other terminal emulators. :)

- And so much more: https://github.com/Uzaaft/awesome-libghostty

Libghostty is really successfully turning into the tide that raises all boats. As it becomes more popular, Ghostty itself gets more performant and stable, and so does everyone else. It is also a vehicle for delivering modern terminal emulation to the widest source possible.

I plan on tagging libghostty 1.0 in the next few months and declaring ABI stability at that point!

Now with LLM's that business model is in shambles and the finance side of it is less attractive with interest rates so high. I suspect that this is actually driving the software engineer hiring slow-down, because these companies were actually doing most of the employing of software engineers, much more than the big tech companies.

The private equity and private credit crowd is one of these groups that are currently running into this issue.

There is no way a growing company($15M to $500M in yearly sales is the sweet spot for these private credit companies) can compete with Google on debt terms so they have to raise their rates to compete suddenly, not because of all rates rising but strictly because we've just seen a huge debt issuance from some very solid credit companies.

We haven't seen a shock like this in 20+ years I've followed the credit market where a small group of companies issue so much high quality debt in such a short period of time, note there is also a lot of less high quality debt issued in this same time:)

So much has been issued that credit funds have run out of appetite to acquire Lowe middle market debt at their current spreads. This is just the way markets work so no one is crying in the casino but it is a big shock to the credit markets.

heck there was concern that the US government may have trouble issuing 5-10Y debt at rates they are used to because of the amount of GOOG, META,etc debt that is available at better terms.

Its a wrinkle to the markets that no one really predicted as a knock on effect of AI's growth.

This is all the back drop for BDC(Business Development Companies) and other private credit issuers who normally loan money for 5-7 year terms.

Those cheap 2021 loans to these small private companies are now coming due into this market with the additional head wind that the largest single sector that issued debt was venture backed software companies, who now face the head winds of the macro factors described above and the additional factor of AI commoditizing alot of hte software that these companies are building.

Private credit and equity are going to have a rough couple of years.

Things to watch out for are lack of IPO's and exist from venture/PE backed companies and a proliferation of continuation funds that kick the can down the road.

On the plus side, its a great time to be a credit trader, but not necessarily a credit issuer:)

I think I'll keep my Framework 13 Pro preorder.

I worked in early AAA in the late 90s, and it's true that games were often quite rough around the edges. But they were also often biting off WAY more than they could chew, inventing genres and innovating like crazy, often in messy ways that didn't always work. But that meant players were often also routinely and genuinely surprised by the experiences they encountered in even big budget games.

He lists a bunch of quality of life and balance issues that were fixed in Diablo 3 compared to Diablo 2... but, well, Diablo 2 was obviously _important_ in ways that Diablo 3 wasn't. It's a much greater game.

And lurking in the background is the hard fact that often the progress with software (certainly game software) entailed turning to much better software engineering standards, using much better external libraries or game engines with much larger teams, and relying on much more process to capture and filter all those rough edges... and along the way, you've built a much, much larger team with a much larger burn rate, and so all those experiments and innovations become much more dangerous and need to be corralled and controlled and mostly stifled, and as the cost of all this to build skyrockets your entire game design _has_ to focus on your business model and your microtransaction story and your live services first and foremost, because you've boxed yourself into a corner where that cash flow issue is the central fact of your game / software design.

(I experienced this evolution first hand in the FPS space)

All of which is to say, I think there are multiple different axes at play here for evaluation, and it's probably fair to say that software _undefinably_ was much worse in the past along some of them... but not all of them. And there are probably deep tensions and tradeoffs between some of those axes.

I think this entirely discussion really is muddied, though, too, by the fact that hardware got radically better over this entire arc as well, and so many programming techniques in the past that made software more brittle or crash prone hinged on memory and computation optimizations techniques that really had nothing to do with the software culture of then or now. It was just a response to the realities of the hardware.

I'm sure people will objected but I'd argue WebGPU is easier to understand, has better errors, has no state machine, works C++, Rust, JS/TS and other languages, and closer to modern GPU programming if you want to move to Metal or D3D12 or Vulkan later.

Further, OpenGL (and ES) are a nightmare of bugs and incompatibilities students might run into that WebGPU fixes.

That said, the article mentions Apple platforms where you can use ANGLE if you happened to have students stuck there.

Computer science used to be tiny. In the late 1970s, AI was a few tens of people at MIT, at Stanford, and at CMU, with a few smaller groups elsewhere. I met about half the greats of my generation, as did everyone else who was serious. There were almost no textbooks in the early 1970s. When Knuth's Fundamental Algorithms came out, it was great to finally see what had previously been folklore on mimeographed sheets in an actual book.

I was involved with Autodesk in its very early days. Even went to the founding meeting. The feeling was that PCs were finally getting big enough to be interesting. The founders were all mainframe programmers, and the people who came up from 4K memory on an 8088 had no idea how to organize a big program. But mainframe people did.

There was a vast emptyness in the product space ready to be filled. Products considered were a window manager ("AutoScreen"), a text editor (Kern Sibbald actually built that), a database, and a CAD program. None of those existed yet as successful products. CAD was picked because the productivity gain was so huge vs. paper and pencil drafting. At the time, the state of the art in drawing editing was drawing on mylar and using an electric eraser to delete. Then you ran the Mylar master through a big copier to make blueprints or whiteprints. A CAD system on a PC was a very easy sell.

So all you had to do was make it work. No need to overcome the noise level of a thousand similar "apps".

You have the guy who wrote code in the 90's, became a dev manager, maybe is a PUM now, or still a dev manager, and basically "does emails and attends meetings", goes to the "daily triage" (which has a whole bunch of people in the room), is a member of the "war team" (they probably don't use that term any more). All these people are not "doing" anything useful and there are a lot of them at MS.

This used to work out fine, because the way the August reviews always worked (stacked rank etc.) it's very much a favorites game, "who you know". So you get a pass.

But suddenly there's a reorg and new management comes in (or a flattening; no more dev leads with 2 reports!). And they have their own set of dev leads, dev managers, etc. they want to install. So the existing management gets kicked out, those who still have buddies in other orgs will transfer there (but this gets much rarer now, as most friends from 'old Microsoft' have retired). But the "does emails & attends meetings" guy is out of a job (and is going to cry "age discrimination" because he's 50+, but the reality is he hasn't done a whole lot for a decade, and has zero useful skills now).

PMs; there are WAY too many of these, with their entire parallel org of management. There used to be almost as many PMs as devs back in the day. Other companies don't have anywhere near so many. Most of them can go; back in the day a lot of them would go to the dev's office, ask about a topic someone asked about in email, the dev says "Oh I can just reply to the thread", PM says "Nooo... let me do that" (PM wants to be the one replying to be 'visible' even though they are just a glorified status collector/forwarder).

There ARE a lot of incredible people there, and the best people I worked with at MS are just as good as the best I worked with at the FANGs. But unlike other companies (FANGs, startups) there are a LOT of a really bad/useless people there too. A lot. Although I have no faith that MS is going to lose THOSE people in this buyout.

Engineering discipline is absolutely non-existent at MS. Everything is date driven. The source tree (Source Depot) gets locked as of X date and only "test fixes" can be checked in after that. But the next ship vehicle is months away. So people keep checking in regular code as "test fixes" after the branch lock. One time it got so bad that the build was actually broken for the entire day of the branch lock, and people were force submitting in the evening (everyone expected whoever broke the build to fix it, but... they didn't, so force submit!) to make sure their changes got in before the lock.

Random unrelated funny MS story: The wiki that everyone in Bing used, got wiped out, because it was running on a machine in someone's office that got wiped when they left. We had to have everyone run a script to salvage what they could off everyone in the org's browser cache.

So there is some truth in the claims of the MS President.

Let this serve as a reminder to vote! Removing this dementia addled criminal and his supporters from power is the patriotic thing to do if you care about this country and its future.

Terra and Luna are dirt cheap and good enough for 50% (or more) of the typical junior dev's workload, and great for routine work. Sol and Astra have so much written about them at this point that it should be obvious what those are good at. The large/gigantic models for things that need full intelligence and slightly more branching thinking.

But that's not the point; the point is understanding these things better by seeing how they are made (or grown, if you prefer). Building something at a level further down the stack than you will ever go forces you to confront details that shape the layers above that you do deal with. This is one reason why Crafting Interpreters is such a beloved book here on HN.

LLMs are now at the core of tools we all use everyday... don't you want to know how they are made? Maybe not everyone has this itch to understand the platforms and tools they use, but I sure do, and it's helped me over and over again. Sure, LLMs aren't like operating systems or compilers. They're a quasi-organic mass of billions of parameters learned during training and honestly nobody (not even the mech interp folks) really deeply understands why they work so well. But that's no reason to completely give up and there is still a lot you can learn.

Take RLVR ("Reinforcement Learning from Verifiable Rewards"). This is the training technique that made LLMs get so good at programming and math in 2025 and 2026. A lot of people who confidently talk about LLM capabilities either haven't heard of it (and thus end up parroting out-of-date factoids about LLM training derived from the pre-2025 era) or have only a very dim idea of what it is. In this course you actually do RL training of an LLM to get it solving math problems better. You're right down in the details. It's interesting for its own sake but it will also give you a grounding for discussing questions such as: "why did LLM writing get so weird and bad in mid-2026?" and "why is it so hard to train LLMs to be good at things that aren't like math and programming?"

Anyway, Sebastian's two books are great and highly recommended; I worked through the first one a year ago and it was one of the best investments of my time in all of 2025. If you prefer video for learning this is great as well (although I'd still say - try the books first).

While I do believe that selling to DigitalOcean was a mistake ultimately, I'm concerned about the long time viability of this move considering the current landscape of web development.

Don't get me wrong, I think CSS-Tricks is a great resource and it has helped me over the years, but if I was in Geoff's shoes right now I'd be pretty scared.

Look at TailwindCSS for example, they had to reduce the size of their team and ended up joining Shopify to continue surviving. It is hard out here these days.

I don't want to sound too doom and gloom, so I wish Geoff the best of luck and I hope he succeeds.

ofcourse its not great, but afaik we still can't do much better. Perhaps if universities and their medical centers got proper budget they could grow humans and human parts to test on. theres many things like artificial wombs and 3d printing of tissue etc. but afaik these will be hugely expensive and way outside their budget.

Or does peta prefer only big pharma to be able to do research and sell simple medicines for ludicrious prices (as we've all seen them do already).

Wonder what other people's take on it is.

i also wonder why they dont fight for human rights instead of animal rights. i mean, sure animals are essential, and sentient, and know what is suffering and incarceration, but so do many humans around the globe, even in 'developed' countries.

maybe they should put them in a childrens' hospital full of kids with incurable diseases and ailments -_-. i've been there and seen _that_ and its a ton more depressing and horrible and a few monkeys suffering (which is still horrible but magnitudes less i can tell u).

try looking one of those kids in the eyes and tell them you are advocating for something that will prevent their chance of a cure how slim it might be. im quite sure u need to be pretty unempathic to be able to do that to say it lightly....

this kind of advocating causes litterally violence to some of these researchers... completely bonkers.

Maybe this is how early Google people felt or how early OpenAI felt. I don’t know. They do have an embarrassment of riches, but I am sure they also felt they were doing great things.

I also worked at modern-ish Amazon and it sucked. Horrible office environment, tedious tasks and overall sense of oppression is what I felt. It is such a huge company, maybe there were pockets of joy.

My current job is back working on hardware products and the software that runs in them. It isn’t quite the same feel as 80s computer companies, but some of the joy is there.

I’ve worked a 2010s tech office setting, but that was from what I can gather quite a different beast than working in a tech office in the 80s and 90s. The things we worked on seemed so much more frivolous than what our counterparts did back then and culture was more focused on becoming a unicorn or getting bought up than creating something great.

I dunno, I look at records of those times and wish I could’ve been there and worked on something meaningful instead of bouncing though SV working for companies chasing dumb trends that turned out to be dead ends.

It's not a great position to be in and it will still take a long time for the compute gap to close. In the meantime, Anthropic, OpenAI, and Google are essentially unchallenged.

[1] https://www.konstantinpilz.com/posts-publications/chinese-st...

Anthropic thinks it is. The got the demand / datacenter match right in Q2, got great margins, and committed hard.

They've got:

- $114b committed to Google

- $110b committed to AWS

- $161b committed to Broadcom

- $120b committed to SpaceX

If the party continues, they'll go from $50b in revenue to $500b and be fine.

But if they don't, it's going to hurt. And that's their only choice, because otherwise, they end up with a frontier model and 50% of the capacity and get eviscerated on either availability or last-minute datacenter leases.

It's a big business, but it's not an easy business.

Birdwatching can still be enjoyed in most places without needing to travel far. Birds are less active in bad weather so you don’t need expensive outdoor gear, and in many places there will be freely accessible nature reserves with hides in which one can sit in shelter to watch birds.

The highest quality glass is going to get pretty expensive (camera lenses can get to the region of $20k; binoculars $4k or so) but there are also much cheaper options and often these can be quite reasonable: you don’t need to take photos of birds; a superzoom point-and-shoot can work great; more expensive binoculars make a bigger difference in low light and if you’re young (older people’s pupils won’t dilate as much so bigger objectives at a lower magnification won’t be a huge help).

I feel like mostly you’re observing that rich people spend lots of money on their hobbies and not that opening your eyes to look at birds is some expensive exclusionary activity.

A lot of, "birdwatchers" I see travelling to remote parts of Scotland and other places in the UK are generally wealthy retirees, with nice kit (Rab clothing, telephoto lenses, VW campervans).

Cognitive decline is shown to impact disproportionately on less wealthy people. So is there a trend with economic wealth and outdoor hobbies like birdwatching? I'd hazard a yes.

On a side note, the Merlin bird ID app from Cornell University is great.

I recently watched Toy Story 5 on a long flight. I almost felt it should come with a trigger warning - I thought it was a spot on, brilliant yet inherently depressing commentary on our current relationship with tech.

(Minor spoiler alert) One of my favorite scenes was when a slew of toys were running through a house with a family inside. Normally in the Toy Story world when a human see the toys they need to "play dead", but since all the humans were glued to their screens they could run by unimpeded.

64dd on original 3DS

Local & Online Multiplayer

QR Code sharing for Custom Courses

I hope ya'll check out the authors notes, post-morterm, and give it a try on your 3DS.

I claim no great historical authority here but, since I grew up mostly within sight of the hillfort, we were taught about its history at school to an extent that, at the time, felt unrelenting, cruelly repetitive, and utterly tedious. In reality though, it probably lasted for a single term at the fag end of 1985 when we learned about the Celts[0].

But what we learned was that General Vespasian captured Maiden Castle after a brutal one day battle in AD 43 in which Maiden Castle’s Celtic defenders were slaughtered in what amounted to a one-sided rout due to the superior weapons, tactics, and discipline of the Romans. This is what I remembered, but I wanted to look it up to ensure my facts were straight.

Welp… TIL because subsequent investigation and analysis has shown that it didn’t happen at all so, no, dear reader, my “facts” were not straight.

There never was any big battle.

There might have been the odd smaller skirmish during the Roman period, but the bodies we found that were all supposed to have died in the big battle actually span several generations, dying at different times over a period of a couple of hundred years, and the evidence of violence in their deaths was in the vast majority of cases inter-tribal and occurred before the Romans even showed up.

Maiden Castle had already fallen into decline by the time the Romans arrived and was mostly abandoned for a long period until a temple (that you can still visit the ruins of) was built early in the 4th century.

So the reason the battle of Maiden Castle doesn’t appear on the site isn’t because the site is carelessly built AI-slop but because no such battle ever happened.

I have been… disillusioned. Hats off to the human behind the AI that built the site.

[0] I remember this with such precision because it was the first period of history we learned about at middle school, and matters didn’t improve significantly, because next it was the domestic matters of the bleeding Victorians, and lots of episodes of “How we used to live”.

Now my agents mainly write code under my direction and I review it either in the harness or on GitHub.

I hadn’t opened it in ages and when I did need an IDE something like Zed was more than sufficient.

If you ever get the chance to go, please go! Especially if you're an American I recommend it. It is so different than any other conference I've ever been to in the US.

It doesn't have a lot of sales pitches, the technical content is top notch, and overall it is really fun and interesting to see what everybody is talking about and working on.

I do feel like I should be thankful to be alive right now though. I witnessed the evolution from Kirby on the Game Boy to Half Life 2 to VR to LLMs. The tools we use today were science fiction only a few years ago.

I just wish society was in a better state to make this tech a greater good instead of a tool for advertisement, warfare and wealth concentration. I feel a deep sense of loss because of this. 20 years ago I had no doubt that technological progress was an obviously good thing, that it would improve the lives of everyone. Do I still believe this? I'd like to, at least.

Jonathan Blow is a glaring exception to this idea. In order to make his latest game, which has shipped and is great, he wrote not only his own engine but his own programming language and compiler to build it in. Zero use of AI.

The project would've been impossible for me without LLMs. & even with them there's an argument that this project is pushing complexity limits too far. But at least with ETL there's a clear feedback loop to build on

> Brazil’s government announced Friday that annual deforestation in the Amazon fell to its lowest level since records began in 1988 [. . .] During Jair Bolsonaro’s presidency, from 2019 to 2022, Amazon deforestation surged by about 60% [0]

This highlights the greatest difference between these two candidates (imo), one wants to preserve the 'lungs of the Earth', the other wants to rip them out for profit and give the world a middle finger on their way out.

[0] https://apnews.com/article/brazil-amazon-deforestation-lula-...

https://pkg.go.dev/git.mills.io/prologic/go-gopher

1. If AI would bring great productivity but all that extra value is captured by closed-source model providers, then everyone would switch to open-weights models + bare metal servers, which would non-tech companies to capture most of the value. Your argument that there is great value in the use of AI, but it's all captured by OpenAI and Antrhopic falls flat on it's face here. The market would adapt quickly.

2. OpenAI and Anthropic are still losing money. Your argument is that 100$ of extra value is produced, but the closed-source model providers capture 99.9$ of that because they are "rent extractors". If you're right, and they capture 99.9 of the value created, that's a huge issue because they spend 150$ to create that value. If even after eating the whole (value creation) pie they're still unprofitable, that's the worst news for the AI industry.

The domains are missing from the public suffix list. This impacts how cookies are handled by the domain names, with concerns the browser will allow software to set the cookie on the shared domain name instead of the subdomain. This can be a security problem. Getting added to the PSL may not happen: https://github.com/publicsuffix/list/pull/79

It can be difficult to get a free HTTPS certificate for your subdomain. Let's Encrypt applies per-domain name rate limits and these domain names see lots of issuance requests. There's a process to lift those limits, but last I checked FreeDNS didn't use it. You can still, of course, pay for a certificate.

There's tons of abuse on free domains so you may find that your domain is blocked on various lists.

Here are some compelling alternatives I use now, which manage these issues better:

- https://messwithdns.net/ - Exceptionally well-made DNS console for learning DNS

- https://open-domains.com/ - Uses Cloudflare, great Discord group

- Even more: https://free.hrsn.dev/#/?id=domains

I find that banking apps deserve that extra bit of anger pointed towards them. (It’s also increasingly hard to find an ATM in many countries. What are banks up to?)

Here is the (much, much higher quality) original reporting that this terrible ny post writeup is based on: https://www.theguardian.com/environment/ng-interactive/2026/...

"There are the live video compositors everyone complains about and there are the ones nobody uses."

But really, it is great to see a piece of open source software as the default choice in a field where they tend to suffer, desktop applications.

I want my type checker to check types and to scream at me when it can't. I don't want it to do "gradual typing", if it is unsure about what type something is, I want that to be a failure instead of it modeling the unsureness and not showing potential errors. We live in the age of agents, having to add an extra annotation here and there is not so much of a problem any more.

Meta's Pyrefly largely does what I want. I don't quite love Pyrefly either; I don't like the fact that it edge-cases and gives monopoly status to some popular packages (Pydantic, Django, attrs) instead of offering generalized mechanisms that less popular packages could also utilize, but purely on a type checking front, it's excellent.

I recently ‘rescued’ [0] one of my favourite video game series, No One Live Forever, which is kind of lost to IP hell. It’s been an enjoyable experience, I learnt a lot, and it’s feels great to have revitalised the game.

0: https://github.com/nolffan

We'll publish the technical details of these hacks over the coming weeks. Writing these articles takes time, and we've been busy with a PHP thing [1].

In the meantime, check out The Great Glyph Grift [2] for a glimpse of what can go wrong when iOS parses PDFs.

Just in case you wonder, we've disclosed most of the bugs in these exploit chains to Apple and will disclose the rest. So far in 2026, we've reported close to 100 bugs to Apple. We don't accept Apple bounties, and we're not being paid for this work. We don't sell bugs or exploits, and we have no sponsors funding this research.

We do this because it's fun, intellectually challenging, and we want to help make the iPhone safer. We also have a business to run, so we do need some marketing points.

[1] https://calif.io/php2shell

[2] https://calif.io/research/the-great-glyph-grift

I've been working on my blog lately, and having only a few posts up and no readership, I'm now blown away by all the comments and mails and support I'm receiving. Thank you so much everybody!

It's a privilige to live in Germany, where I have public health insurance, and access to great doctors and staff. I'm tapering off the cortisone right now (I'll be taking it for six weeks in total, so I dodge a lot the strain it can have), and am now on another medication. I don't feel comfortable sharing my condition / medication due to me being on the job hunt, but maybe one day. Thanks for all the pointers, too.

I think part of the success of this message is that it’s so obviously silly that people feel silly questioning it.

But that doesn't make it a failure. In the era where we have:

a) an AI hype cycle where the benchmark of "successful product" is ChatGPT, Claude Code, etc. b) a social media environment that only rewards the most-engaged content

It's easy to declare something "fell off" because it didn't immediately revolutionize how everyone does their daily work.

But by that measure, nearly everything is a failure, even if it creates great businesses. That's especially true for infrastructure that runs in the background.

Meanwhile if someone lives in a city they'll get loads of results.

EDIT: actually this is just a wrapper around wikipedia's nearby feature I think. I tried a couple of places and they give identical results.

https://en.wikipedia.org/wiki/Special:Nearby#/coord/15.90721...

"Several years ago, when I was ill, Heinlein offered his help, anything he could do, and we had never met; he would phone me to cheer me up and see how I was doing. He wanted to buy me an electric typewriter, God bless him—one of the few true gentlemen in this world. I don't agree with any ideas he puts forth in his writing, but that is neither here nor there. One time when I owed the IRS a lot of money and couldn't raise it, Heinlein loaned the money to me. I think a great deal of him and his wife; I dedicated a book to them in appreciation. Robert Heinlein is a fine-looking man, very impressive and very military in stance; you can tell he has a military background, even to the haircut. He knows I'm a flipped-out freak and still he helped me and my wife when we were in trouble. That is the best in humanity, there; that is who and what I love."

https://en.wikipedia.org/wiki/Philip_K._Dick#cite_note-32

I feel GTA 6 will be the last mainline GTA that we run on our local systems.

An editing pass to make the prose less "punchy" would be great.

This is in asymptomatic patients, who already have a great estimator for 10-year heart disease mortality risk in the coronary calcium score which isn't even obliquely mentioned in their abstract. Edit: the paper shows no correlation with calcium, either

https://www.medrxiv.org/content/10.64898/2026.10.19.26364786...

In one (which I can't find) he mentions a Japanese musician coming to LA to record with the top LA studio guys. It is very common to play to a click track, and one song just wasn't gelling. The studio guys said, "Let's try it one time without the click track." The Japanese musician said, "We're allowed to do that?" Anyway, they went ahead and the song came to life. The musician literally cried after because it felt so good to not be focused on tracking the click instead of being in the moment making music with a bunch of other humans.

[1] https://www.youtube.com/@lelandsklar6363/videos

On a more depressing note, wasn't this always a thing? I remember being in 4th grade when a classmate brought in something called "Mise-o-jou", which was being run by the Quebec government.

Step 2 is instead of waiting to be caught, the instigators should issue a press release blaming cunning rogue agents as part of their research, call no felonies because they’re working for the greater good of eliminating a shitload of jobs, say the real problem is a lack of regulation. They could even return the money if it looks like they’re going to get sued, and collect some sweet, sweet reflexive investment money.

But I question its utility on other platforms, because a lot of the things that makes PowerShell cool on Windows doesn't actually work in Linux or macOS.

Eg: Get-Service (and associated cmdlets). Very handy cmdlet that I use all the time on Windows. On Linux, you'd expect that it'd work with systemd units, but no, that command doesn't exist on the Linux version. Same with Get-NetAdapter, or Get-NetTCPConnection, or Get-Disk and so many other handy cmdlets that I use all the time on Windows, are all missing.

Yet, some cmdlets do exist, like Get-Process or Get-Uptime etc, so I don't understand why many of the commonly used Windows cmdlets haven't been ported over.

Another annoying quirk is how PowerShell on *nix hands off some of the built-in aliases to their native versions. Like on Windows, I always use 'ls' because ain't nobody got time to type Get-ChildItem. Unfortunately when I run this on *nix PS, it runs the native version. And this is a problem because when I pipe 'ls' into a for loop or another cmdlet for further processing, it naturally, doesn't work. To make things worse, this native-or-alias decision making isn't even consistent. Like 'pwd' for example actually runs the PowerShell version, whereas 'echo' runs the native version.

Basically, as much as I love PowerShell on Windows, on *nix it is very annoying to work with. I would much rather work with native *nix shells and scripting languages. And if I really wanted the structured output of PS, I would run Nushell instead, which is far more consistent and useful than the half-baked product that *nix PowerShell is.

First, Apple does not follow the same rules they set for developers.

Second, Apple does not treat all developers equally. We have functionality we’re unable to offer to iOS users.

The same feature, including technical implementation mechanism, is available in Claude and Gemini. It’s unfair.

Going forward, I hope Bioware's people land great new jobs, and here's hoping some of their IP gets picked up by other studios. Mass Effect in particular is a fantastic creation that a lot of people would love to spend more time in.

I absolutely loved being here today.

So many brilliant minds screaming and hating together, trying to survive the tide.

There aren’t many places like this left on the commercial internet, especially in this brave new AI-driven world.

I felt a great disturbance in the Force. But in a good way.

Together, we are strong. May the Force be with us all!

New Mexico decided to go it alone and gets $25 billion.

Great timing for the AG who’s up for reelection on November 3 (even though it was the previous AG who made the call).

Good tooling, modern language features (C# 15 has the long awaited unions, closed hierarchies), great debugging experience, etc.

[1]: https://www.reddit.com/r/dotnet/comments/1x7wsgb/rust_just_l...

Instead they seem to think that AI can someone work around what is essentially laws of nature.

1: Benzo's, as they call them, reduce inhibitions greatly, and also reduce short-term memory. If you take a 1 MG Xanax with no tolerance, and you have a bottle of 10 more pills, somewhere within the next 12 hours, you will take all 10 of those extra pills. The "High" has a very "I don't feel it" quality, because all that's really happening is your inhibitions are vanishing. So, geez, I don't feel it, I'll take more! Keep in mind, a 1 mg Xanax is usually supposed to be cut into 4 doses.

2: Driving on Benzo's almost always results in a "Americas Most Dangerous Police Chases" type behavior. The driver will likely have taken, now, 11 mgs of Xanax, and will awaken in a jail cell 2 to 3 days after the car incident, completely unaware of what has transpired. Guaranteed they fought the cops 100% of the way to that cell, too, so they don't know where the bruises came from.

3: A real Benzo fiend "blacks out" all the time. A common theme in this subreddit would be "I took 1 MG, had a 100 pills left. Woke up 1 week later with pictures on my phone of myself and 3 other people having an orgy. No pills left. I don't remember any of this." It's astonishing how much people can down in such a binge. Seems it's hard to OD. HOWEVER.....

4: If you are addicted to benzos and you quit cold turkey, you can die. Seizures, blackouts, flu-like symptoms. Its BAD. This is not Heroin. You have to slowly taper off of them, like you were decompressing from a deep ocean dive, over weeks, if not months. Or else you DIE!

5: You can go online right now and get a prescription for this stuff from some disgraced doctor through a startup like those Viagra, ADHD, or GLP1 pharma mills.

6: DON'T EVER TAKE A BENZO, EVER. Maybe 1 before eye surgery where they can't anesthetize you and you leave without any more pills.... Something like that. Otherwise, avoid.

As someone who's contributed to Wikipedia and seen how the sausage is made: that's stupid. Even Wikipedia itself considers wikis (including itself) to lack sufficient authority to be cited as a source.

Such citations could be easily gamed (if you cite a particular change) or are useless (the content you cite may not be there in the future), and are obvious pathways to "citogenesis."

I also love another of their programs, Fanlore[0], a treasure trove of obscure internet fanfiction works, social groupings, and tropes. Sort of a combination oral history + TVTropes-esque fanfic breakdown.

Fanfiction is a beautiful thing and a great way to stretch creative writing muscles that can be helpful by being a bit less of a blank canvas than needing to invent characters and locations. Of course, a TON of fanfiction does invent, liberally, but I've seen people insist they can't write creatively, then make something beautiful when you ask them to rewrite a scene in their favorite show or book that they were disappointed with and boom, they just wrote their first fix-it fic[1].

Also, praise be for AO3 having no news feed/algorithm, and instead an incredibly intricate tagging system. Makes it easy to find what you want, avoid what you don't (see free speech commitment), and feel liberated from algorithmic exhaustion.

[0]: https://fanlore.org/wiki/Main_Page

[1]: https://fanlore.org/wiki/Fix-it

Could a Great Firewall of the U.S. be far behind?

I'll be pulling down models today…

I think everyone has known that inference is profitable overall for the major providers for a while (I'm sure not all user accounts, but in aggregate).

The major providers are unprofitable after you add in the training and research costs and other overheads.

This is why the open source models are such a threat if they become good enough - it will destroy the business model of just providing raw chat/intelligence.

Arguably that's also why a few of these providers are trying to diversify into whole-platform approaches rather than just relying on having their whole thing be a great model.

The company got successful. At some point I had a team, and then multiple teams. Gradually, I had to let go of that level of understanding and delegate it to others. When I was overseeing 50+ engineers, I only had a high level understanding of most the systems, and only got involved in lower-level decisions rarely. All the new code was created by my team, and it varied in quality.

If I had AI tools back then I could ask it to explain things to me that my team did and have a 2nd set of eyes which had a different set of biases and abilities.

That's how I deal with it right now. I ask one AI to gen the code based on a design. I work with the AI tool and see how it builds the system out. I intervene where it starts doing things I don't want it to do. I ask another AI to review the PR after the code is written. I sometimes dig into design choices and ask it to explain them, rate them, and offer other alternatives. I sometimes change things, and occasionally scrap the whole thing and start again with better guidance.

So far I've written more code without an outage this way than ever before with a team of human developers. AI screws up too. But right now, at Opus / ChatGPT Sol and higher models, on average, it makes fewer mistakes than humans did. Any time they slip through is because I am rushing too quickly. In a pre-AI human dev team we would call this a process or management failure.

The more dangerous a decision and the harder it is to roll back, the more time I spend reviewing it. This has served me well when managing people who write code, and seems to work well with AI.

AI is a tool that needs to be managed. It won't do all the work for you, but it can speed you up greatly.

I just approve PRs, because I have to. I don't know what else to do. I can barely keep an eye on all my code.

Actively looking, but job market is rough in Northern Europe. I quit caring about quality in any capacity here, the company has clearly chosen its direction, against my very vocal complaints.

My last interview, with what I thought was a great company, failed when I didn't understand what NLP was when they asked me about it.

> NLP? > Natural Language Programming, it's the future. Claude, GPT, etc.

I laughed, he didn't, looked offended and said it was a serious question. I had thought it was a joke, said "oh you mean LLMs". I tried to answer seriously, of course, but that was my third round technical interview and I didn't get a call back.

Just getting a paycheck and producing mediocre work until I get something new. (Senior engineer, 10 YOE, patents, 3 years at a unicorn).

Also considering becoming an electrician.

I have not really formalized my thoughts on the matter, but it boils down to. It feels great to get the machine to generate a thing for you but it feels terrible to receive a thing from somebody that is machine generated.

The worse part is the paranoia, Now I am seeing "tells" of machine generated content, and it may or may not be. But I have already lost interest.

The real irony would be that it is a hand drawn image carefully animated to feel machine generated.

My great essence mining bot contraption horribly broke when Runescape started randomizing all colors slightly, after making a paltry amount in dollars selling the essences.

[1] https://scar-divi.com/

Reverse Engineering became much easier, it is plus. Finding vulnerabilities also became much easier, another plus. Models are getting better. But where is the new cool software ?

Edit: I will go even beyond OSS. M$ & Goog, who are DEEPLY Vested into AI, have access to the best possible frontier models and large groups of highly proficient engineers. Yet, Windows and Android are as laggy & buggy, as they were 4-5 years ago.

However, most code is corporate CRUD stuff. The world sort of runs on CRUD. But it's rarely particularly fun, or beautiful, or fulfilling. It tends to be pretty ugly because there's rarely time to make it beautiful. So I'm not losing much joy there.

Additionally, even with an agentic workflow, my team still thinks deeply about architecture issues. We believe that well structured code is better for bots and humans alike. So I am still exercising my engineering brain; I'm writing blueprints and supervising the bricklayers now instead of laying bricks.

Does that match your teams' experience at all? Sometimes I feel like folks on these threads are working in a parallel universe where the majority of coding tasks and somehow fulfilling and fun and not 99% tweaking reports, trying to implement arcane and contradictory business logic, or fighting with shitty and frustrating externalities. Maybe people found joy in those aspects. I guess there are all types of people in the world. I mean, I'm a sicko who loves writing SQL, something that most coders seem to hate.

    I've been direct with my team about it being a legitimate 
    feeling of grief and the grieving process that comes along 
    with it. It's been helpful for folks to hear that they aren't 
    alone in that feeling.

I happen to love the process of building software by coordinating agents. It’s an addictive form of flow state that I enjoy a great deal - similar to the flow state I get playing real time strategy games like Age of Empires. It’s a different kind of flow state than writing code, for sure.

But to say that the only people who enjoy building software in this new way are those who hated coding, technology, or problem solving is simply untrue.

AI is great enabler for stuff that you don't want to be good at, it cut a ton of chore type jobs out of my workflow, but for stuff you do want to get better and care for, it's both joy killer, and if you're not careful you will only learn how to prompt, not how the stuff works.

The carpenter is free to exercise his craft as he see's fit, but it's going to take place mostly in his personal workshop, for personal reasons, not to develop homes or run a business in the way he did before.

We can talk for ages about the quality of the robot's work, the maintainability of the resulting home, etc. But it's clear that, at this time, the market and ecosystem are embracing the robots whole-heartedly so those of us used to the old ways are having to rapidly adjust.

I can totally identify with that feeling. It must be how professional mathematicians are feeling as well.

>When people claim that it’s no longer sustainable to write code by hand, they’re not entirely wrong, but only in the same way that it’s no longer sustainable to sleep in a bed once someone has taken a shit in it.

I've resolved to just building stuff for my own businesses and myself for fun. If an opportunity or enlightened soul comes along that wants to benefit from my knowledge, great. But I think it's best to just let people feel the full brunt of what's coming. Sometimes the only way to learn the lesson is to stick the fork in the socket.

And the skills they provide are transferable. Patience, attention to detail, staying on task.

People don't talk about this overlap but it is real, remember that SBF lead Anthropic's Series B with stolen customer money from the Bahamas. Caroline Ellison participated too, in individual capacity.

https://www.privateequitywire.co.uk/ftx-ceo-leads-580m-serie...

This is a great point, and one that really isn't played up because it doesn't fit the whole profit-from-attention narrative that literally every media surface seems to be following.

It almost seems as if law enforcement 'need' these additional tools to keep them busy; find new, more trivial crime that's been hiding behind no one reporting it or being hurt by it.

Maybe if law enforcement resources could be applied to more corporate crime, that'd keep em busy till the cows come home.

https://x.com/JoshuaSWarren/status/2111923039378211100?s=20

My father is mostly on Apple and tries to use his iPad as a PC, ie with files and "the filesystem", he tries to get out of using extra iCloud for pictures (using Proton, or he asks me about Immich, or he wants to use an external HDD, or use Synology Photos and backup). He's mostly on the iPad + keyboard because it's a nicely managed device update-wise and "what I can install"-wise, virus-wise.

But I recent told him: Just give up. Buy (rent!) the storage, you are too depended on me, there is always something with the non-Apple solutions. Yes, your iTunes library is a mess on the filesystem, just don't look at it through anything else than "Music". Yes, Proton Drive eats 400 GB of your iPad storage now, I don't know how to trim that regularly. I'm sorry, it's either Linux and I can help set some things up, or full Apple. Choose. His current mixed, "trying to get best-of-both-worlds", approach is just pure unexpected pain, every time again.

Edit: I also have an iPad, it does work for me, but where my father is trying to use his iPad as a central, primary device (the hub), for me it is a client (spoke) of my own infra. Apps get access to information and I never use Apple services for anything. So indeed, Immich, Proton, Home Assistant, government and banking apps, etc. My iPad is throw-away: If it dies, it dies... I buy a new one, log into all (self-hosed) services and I'm 100% back up and running. Doing the same for my father is also pain, because his Home Assistant will have issues, or his Immich instance, and it's real work to fix, and he won't tell me for months if it does not work and work piles up. So, I tell him to surrender to Apple... It die a little inside.

Nolan Bushnell came to visit Autodesk at our invitation, and we gave him a demo of our 3D version of Pong. Later, Nolan gave me some of my first business lessons at a meeting held at the Thai Garden restaurant at the corner of Page Mill and El Camino. Memories...

I agree that VR isn't the best modality for precision CAD work, although it can be great for visualization. These days I just dictate what I want using a local Whisper instance, have Hemes-webui get my local LLM to write Python code, and then let FreeCAD generate the geometry. For me, that works much better than drawing in midair.

What I suspect Linear doesn't notice; the thing that makes their lives easier; the David Foster Wallace water they swim in everyday - is they're their own customer.

Most software makers have to translate requirements between domains where terminology is incompatible, dogfood to find the micro pain that helps a product become great, etc. but Linear can just... track the build of Linear in Linear.

When the someone behind the software is its customer they'll get loads of judgment calls right first time. Great!

If you aren't your own customer but plan to leverage intuition, plan on being wrong lots- and keep your company and architecture and products compatible with frequent wrongness so once discovered you can pivot to the right thing. No ego.

The opinions are those of the creators and that isn’t changing with AI.

This, unfortunately, takes time, as one has to restart many times because of ideas that turned out not to be as great as one had initially thought. And it results in software that is called "opinionated", and often does not follow what the "industry" does.

There is a common understanding with respect transformers the the output text is the only channel between one invocation and the next. ... so if it wanted to be thinking some 'hidden thoughts' it would have to be modulating its word-choice.

I think this is an artifact of tutorials on attention showing the system proceeding and looking back at words. But the information in the KV cache is not words-- it's entirely latent data, and in the interior layers they KV caches don't even tend to have that much to do with words. Their only guaranteed relationship with the input words is that they're a deterministic function of all the input tokens up to that point.

This means that a autogressive transformer could be having a whole complicated invisible discussion going on in the KV entries which was arbitrarily unrelated to what it was saying. And if it wasn't so it would be much more difficult do stuff like exactly copy a list but then truncate at the Nth entry as the simplest way to do that is to count down internally.

This is not, in my view, a concern or issue-- the model saying stuff that is inconsistent with it's "thinking" is a very high entropy configuration that would not be selected for in training unless you were more or less trying to do that. So not even 'could be' --but always is as an integral part of working.

And the useful safety relevant advice is don't undertake training that trains the model to be two-faced and it simply won't be... because being two-faced is a much more complicated state. You can even see it in humans w/ brainwashing: force someone to say something enough and they'll usually believe it, not just say it even though you were only supervising what they said.

Ironically and unfortunately the only credible kind of training I can think of that might produce duplicitous behavior is RL with "safety filters"--- if you have some dumb nanny process watching what the model says but then reward it for achieving a task, you're doing exactly the kind of training that can produce that mismatch. "To solve this task I need to hack this process but if I do so obviously I'll get punished..."

The greater concern with latent reasoning would be that potentially you'd be operating in a regime where the total computational depth is many times the model*kvcache size hold, potentially in a regime that was never trained at all. It could be arbitrarily far out of distribution. It might even say a naughty word or tell a racist joke!

AI and AI companies have a similar pathway open to them where they will take a lot of information already out there and then gatekeep the synthesis based on that information (obtained without compensation in most cases) as well as to harvest all of the latent knowledge underneath. This is a bad development in my opinion but I'm sure that their shareholders believe it is great.

I can't help but to be a little sobered by the promises of Kotlin and C#. They look great on the surface, but once you have more complicated code to work on they tend to show warts quite fast.

Java is remarkably consistent and really doesn't have any sharp edges. Even with generics and type erasure, I found myself fighting the language semantics in C# way more often than Java.

The XHTML/XML/XSLT tooling was bad, and XSLT was hard to reason about. That meant that amateurs had no way to learn it, and those amateurs went to school and became the workforce.

The web became a winner because it was easy to get something working. And then Google made it a mission to make great developer tools in everyone's hands for free.

I think in 2026 we could build the tooling around a better experience, but it would require thinking about the developer experience first.

It made for a great marketing stunt, it is absolutely useless for a production car. And you can tell the designers realize it but were at the whims of Elon in their interview.

https://youtu.be/JdkIOhFmbKI

We were promised an awd, 620 mile range monster almost a decade ago, and then again 6 years ago. We were delivered a “maybe 310 mile” range, rwd doorstop.

I’m not sure who this car is for, but it’s definitely not car guys. It is the cybertruck of sports cars, and that’s not a good thing…

Edward Jenner invented the smallpox vaccine. Smallpox killed 300 million in the 20th century alone and was eradicated in 1980.

Or maybe Norman Borlaug, who invented a kind of high yield wheat that is estimated to have saved a billion from starvation.

These certainly aren’t the kind of achievements Musk aims for (quite the opposite in fact) but I’d argue they have a far, far greater impact than anything he has even tried to do.

He’s got cooler demos though, I’ll grant you that.

carbon fiber is great in tension, i.e. a compressed air tank, or COPV

carbon fiber is much less great in compression, i.e. OceanGate

Also the OceanGate construction process was bad, even for carbon fiber, because the CEO didn't listen to actual engineers.

  > Tesla doesn't have a great record anyways.

And inevitably two groups will clash, one pointing out how much this deviated from the original, often decade long promises vs another pointing at the identical naming, how it's still a product they like and of course the all time winner "I don't see you do X"...

For what it's worth, I'd file this as a RCS demo for SpaceXGrokSI (or whatever they call themselves this week) above all else and if I were a betting man, might be a prelude to fusing them into SpaceXSuperXTeslaXIntelligenceX...

It's an incredible engineering achievement to build in massive fans and rocket engines into a high performance car. It's truly wild.

It's being sold in two modes. One with cold gas (no combustion) seemingly just for extra forwards acceleration (no flying) and one with real combustion rocket engines, the latter was in the video and is "invite only".

Wild that they devoted so much effort to an economically completely infeasible model. I wonder if the combustion version will even break 3 figures sales.

Even just the cold gas variant seems like a nightmare for production, there's a high pressure COPV in that thing and that's gonna make crashes much more.. exciting, and Tesla doesn't have a great record anyways. But very few of these are ever getting sold anyways.

-

First, it's the tok/s and TTFT from providers. Deepseek blazes through work - at quite good quality - and you don't find yourself waiting for your agents to finish.

Second is the openness. Being able to see the CoT and not being randomly blocked for cyber review. You're never wondering if it's stuck or if it will randomly fail. Combined with the great cost value.. My next subscription will probably be Z.ai

First, it's the tok/s and TTFT from providers. Deepseek blazes through work - at quite good quality - and you don't find yourself waiting for your agents to finish.

Second is the openness. Being able to see the CoT and not being randomly blocked for cyber review. You're never wondering if it's stuck or if it will randomly fail. Combined with the great cost value.. My next subscription will probably be Z.ai

Greatness is a transitory experience. It is never consistent. It depends in part upon the myth-making imagination of humankind. The person who experiences greatness must have a feeling for the myth he is in. He must reflect what is projected upon him. And he must have a strong sense of the sardonic. This is what uncouples him from belief in his own pretensions. The sardonic is all that permits him to move within himself. Without this quality, even occasional greatness will destroy a man.

For some games, like MMORPGs, an offline mode doesn't make sense.

In lieu of that, they need to make the server binaries available, along with client functionality to let you choose a server. They shouldn't be required to offer support, but the nerd community is great at figuring things out. They'll get it working and produce guides.

I have found the best results through a mix of Happy[0], this repo, and computer use to fill in the blanks.

It is _really_ good at the stackup and part config, but creates some absolutely terrible routes. It will snake tiny traces through some SMD pins even though there is plenty of space available outside of the component. When it hits an obstacle, instead of creating 2 45 degree turns, it does 10~30 small ones creating some sort of weird crooked arc.

I have produced a couple of boards for testing this way and they are generally functional.

I have tried KiCadRoutingTools[1] as well which does seem to create routes that make more sense.

Anyone any better experiences with these or similar tools?

[0] https://github.com/aklofas/kicad-happy [1] https://github.com/drandyhaas/KiCadRoutingTools

If you change your perspective from “I’m not gonna spend 500 hours on this” to “30 hours of great gameplay for 30 bucks” maybe it isn’t so bad.

However, so far I can't point to anything substantial that should have been skipped. Almost every level is teaching me something, and those things keep coming up later.

Ultimately, Order of the Sinking Star applies the same formula as The Witness (although hopefully with a little more story, but probably not a lot more), and dials it up to 11. I loved The Witness, so really it's exactly what I asked for, and I think the feeling of the distance left to go is an artifact of a desire for instant gratification.

I think the game is deeply rewarding and fantastic, as long as you understand you're signing up for a slow burn. That's not for everyone, and that's ok.

Yes, nukes would be bad to give control to an AI with no workaround. But so are customer service groups of fraud departments at banks. A kafkaesque hellscape is where a customer or beneficiary of your service cannot get the help they need by policy and by design. Like getting Google on the phone, getting a non-scripted human 10 years ago in a call center, or now getting a non-agentic interaction on the phone. The scripts dominate the interaction.

The IRS, for a time, would cut off your call if their wait system was too busy. That's a critical system in kafkaesque hellscape.

Kudos to zlandes86, you have highlighted WHY these hellscapes start to pop up in systems where the human is removed.

I am an AI user and builder -- I fully advocate for the human-augmented system, NOT the human-replaced system!

We credited Sapling because that's how Meta itself externally refers to the entire system, including the Mononoke server, and we wanted to follow that convention. We will add an additional footnote to clarify this. [edit: this is done]

We were inspired by many things while building Scion, such as the great ideas in Mononoke's Bonsai data model, and didn't intend to obfuscate that in any way. We were also inspired heavily by https://martin.kleppmann.com/2015/03/04/turning-the-database..., for example, which also describes the "derived data" concept using the same words.

We also did not just tell Claude to build us a version control system; we did a lot of research on what would work and not work for our use case and coming from an existing large Git repo (rather than Mercurial, as Mononoke did) required many novelties beyond some of these data structures.

The intent was to give credit to Mononoke without getting too deep in the weeds of comparing and contrasting each point of the data model where we diverged. There are many ex-Meta employees at Anthropic who were inspired by how joyful using source control was because of Mononoke, which deserves a lot of credit to you and the source control team there. We had that experience as a north star from the beginning while making decisions that map to our (different) needs.

Anyway, what are useful features of Steam:

- their dedication to getting Windows games running on Linux (to see off the existential threat Microsoft poses to them, but nonetheless greatly appreciated)

- community forums / screenshots / artwork / guides etc. for every product. Very useful if the game vendor doesn't have its own, and most game vendors don't

- that Steam Sales encourage vendors to offer games at much more reasonable prices (e.g. a few pounds/dollars/etc.) for what is effectively a license, revocable at any time, not a real purchase. £60? Ha ha fuck off, I'm waiting for the sale when it's £5

- also, as far as I've seen, every game distributed can still be downloaded today. People may complain that publishers update a title e.g. to remove licensed music 10 years later... but you can still download the old version with some trickery. And I've never seen Steam actively remove downloads for existing customers, I've only seen games removed from the store so no future customers

- pirates have found that most games, if their only DRM is Steamworks, are incredibly easy to crack. I am very glad Steam has convinced most publishers to use that alone and nothing worse, and I curse publishers who insist on their own DRM, their own launcher, etc. If Valve goes tits-up, I have a lot of confidence most of peoples libraries will still be playable with a little bit of cracking.

There are other useful features, even if Valve is getting rich and fat off selling gambling and day-trading to teenagers.

Like immagine if the rhetoric of Turkey was constantly talking about expanding to Greece and how the land of Greece is historically the Turks. Obviously that would make any Greek uneasy.

I don't know why anybody would. It takes all the lessons we learned about UX from user testing in the late 90s/early 00s and throws it all away to produce something that looks "clean" and "sleek", at the cost of being useful. Discoverability? Who needs it?

EDIT: Though I do enjoy skeuomorphism, I do think the TFA goes a bit too far.

Turns out when battery voltage gets low (but not yet dead) this can happen in a circuit. I swapped out the batteries and our ghost never came back. Not surprised to hear it happening in a store where stock just sits there for a long time.

Pretty creepy though to hear that when you are lying in bed

I sat there for 5 minutes reading them and then I tried to type something, and realized I couldn't think of a single thing. Just a bad day. I love that you made this, though. It's inspiring.

Once I figured out that this was a skill and not just something people that are not me are born with, it made things a lot easier. Went from making up all kinds of stories in my head about myself and others to thinking about each interaction as practice and realized lot more people were like me, thinking I'll be invited somewhere or talked to if I was liked or was interesting enough; waiting for someone else to do the inviting or talking first. I just remind myself that a lot of others also want to be invited or talked to, and that reminder greatly reduced my fear of rejection.

My mom was diagnosed with stage 1 cancer in 2006 in Idaho. I was a PhD student at MIT at the time and a friend in lab helped me get an appointment at Dana-Farber. I thought it was probably overkill (it was a common cancer) but we did it anyway. They said everything in Idaho was standard of care but suggested one extra test, which Idaho team then reluctantly did, which revealed distant metastasis (stage 4! ), dramatically changing her treatment trajectory. She lived another 15 years (I miss you Mom).

Asking for a second opinion sucks. A lot of us are preconditioned from birth to trust our clinicians. But you're going on a journey and your life (or the life of your loved one) is on the line. Insurance often covers it. Your doctor will NOT be offended.

Think of it this way : if you were deploying code to the mars rover, and your colleague asked to do a code review before you pushed go, would you be offended? This is even more important than that.

I think a big part of the problem is that middle-aged men are at the apex of their expectations to be successful in the eyes of society, and we are living through a period of time where success is very hard to define, much less attain.

There is a name for this in sociology: anomie [1]. Émile Durkheim coined it in his book Suicide, written not too long after the Industrial Revolution.

In times of great change, the set of norms and rules that we take for granted as an organizing principle of our lives get uprooted. It's as if all of the pieces and tokens and rules of the board game we played our whole lives are swept off the table to make room for a new game, but the new game has not been set up yet. In that interim, we all feel a sense of unease. It is our turn to move, but we don't know the rules of the game or what moves to make. We feel intense pressure to play the game well, but don't know how.

In the past, these transitions have ended and society has generally returned to some sense of stability. A new board game gets set up, new pieces doled out, we learn to play the new game. I hope we get to that point with AI soon, but the revolutions seem to keep coming faster and faster.

I think for a lot of men, it feels like just at the time when we should be able to rest on our laurels a bit after having worked so hard to play the game society told us to, the rules have been ripped out from under us and any sense of security going forward is lost. We feel too old to learn the new rules of the game in time to play well, but not old enough to be able to abstain from playing.

It's a psychologically very difficult time.

[1]: https://en.wikipedia.org/wiki/Anomie

And this isn't even to mention money, and what my future is looking like, despite having every possible intrinsic and external advantage in The Greatest Country on Earth.

I've had one same-aged friend kill themselves, and I try to be attentive to the signals being put out by everybody in my circle.

No other big-tech company goes to such great lengths to deanonymize users and make its services difficult to use, especially during account creation.

The requirement to have a mobile device to scan a QR code to verify that you're human. Reverse SMS verification, where you have to send an SMS rather than receive a verification code. You think of signing up using a VPN server in another country? The phone number you use to send the SMS from must also be from that country.

Not being able to watch YouTube videos when using VPN - without signing in to a Google account, all under the guise of "bot protection."

Gemini not working in 3/4 of cases when using VPN.

It's essentially almost impossible to create a Google account without giving up your identity, and your phone number often leads directly to your real-life identity.

I'm not sure what happened in recent years, but they definitely switched their strategy to maximize data collection and make sure privacy-savy users are pressured via technical solutions to reveal their identities.

PS. The first company federal investigators often go for data is often Google

My mental health improved quite a bit when I realized I needed to identify and prioritize more restorative forms of play. That did mean allowing my childhood passion to become just a day job. But that's OK. I honestly wish I had let it happen sooner. What got me to where I am now won't necessarily get me to where I want to go next.

If that's not how things work for you and programming for fun is genuinely restorative, great. I just wanted to put my own experience out there because tech's hustle culture can make it hard to realize that advice like this isn't for everybody. Especially for those of us who might have a touch of alexithymia. Which seems to be pretty common among professional programmers.

That could be the maximum I'd like out of font to have "proper" greater/less-equal sign.

For me the great app platforms:

* the terminal

* the iPhone/Android

* the web

And suddenly, the web is brought to the desktop and it is cross-platform. Everything is here on my Linux. It is the Year of the Linux Desktop and everyone is complaining about web platform. But it is web platform that has brought it here. Praise be the forces that made this happen. This man has a great website, and if he has made Electron then he is always welcome to me.

Making your own framework is great if you're in no hurry at all - maybe you want to learn, or maybe that's part of what you're selling. For everyone else, use the approach that gives you the most leverage.

This is a clear step change, I'm genuinely amazed.

Its answers are to-the-point, logically structured, clearly written. Like a really well made mini essay on your question.

It also really likes to build little simulations / models that you can tweak the parameters on. Amazing to explore how different assumptions affect economic scenarios, for example.

It's also great at picking (sourced) illustrations.

I'll be using this for everything that isn't coding from now on.

At the same time, OpenAI had a disastrous dev day that led a lot of users to jump ship to Claude.

Couple weeks later, Gemini-4-argon is still unavailable on anything less than the ultra tier, which means there's no natural funnel to get ppl to pay for the more expensive ultra plan other than "trust us the model is good".

If Google had argon broadly available earlier this month they would have picked up a ton of new users.

Here's low, medium, high. Even low is better than pelicans from any other lab: https://tools.simonwillison.net/markdown-svg-renderer?url=ht...

low cost 10.185 cents, medium was 11.098 cents, high was 17.166 cents.

The costs are similar because it used more output tokens than reasoning tokens for all three pelicans, even for high.

But their great idea to deal with the shortage is intermittent availability of power. You won't be able to wash your clothes when you feel like, you will have to adapt to when there is spare power, like in developing countries. This is just a regression in technological progress.

I saw the same messaging in France where the current president started his mandate with a grand programme to switch off a third of nuclear power plants.

If anyone thinks adults are in charge, let alone engineers, you are greatly mistaken.

I watched their video demo with the family getting ready in the morning: there are some interesting ideas here but imo if this doesn't get reasonably full access to the App Store then it's a no-go for me. Like: FaceTime while cooking, sure, but those calls tend to happen on Messenger and Snapchat for us. Music in the kitchen, sure, but I run Plexamp. My kids are always stealing my iPad to run the Toybox 3D printer; this would be a great thing for them to use.

And so on.

My house has a bunch of spots on the wall where an old intercom system used to live: it'd obviously be a budget buster to buy a bunch of these to fit in those slots but even if you did, you'd need Apple to provide an intercom solution (a la the Watch's Walkie-Talkie?) or you're...out of luck.

> Software teams are capturing a lot of the speedup in the form of more leisure or quasi-leisure time. (To greatly simplify: software teams are offering 2x speedups to their companies. They're then using tools that allow them to work 10x as efficiently, and taking nice long coffee breaks.)

That is just an insult. With AI we are working more than ever, either because of "you should be faster now" or because half the people are getting layed-off ...

The current slew of excellent LLMs will probably cause a decades long dip in human cognitive ability.

If social media on smartphones had an impact of ability of students to concentrate, LLM AI will likely have a much greater negative impact on education.

There are trillions of dollars of investment pushing for humans to trust whatever an LLM says, and very little money behind encouraging people to tackle hard problems themselves and train themselves to think.

I have to say honestly, that AI looks like an apocalypse for human thinking [ Im glad Tao is warning people, I just dont think many people will listen ]

This seems to be the case with most industries that LLMs are impacting, coding, data input, legal etc.

Whilst it's great to get the time saving from not delegating to a junior (LLMs are much faster at the basic level 1 work) there's going to be this big knowledge gap that you really only gain from actually doing the mundane boring shit.

A lot of this _should_ be automated away, but we really need to think about how the next group of people are going to learn the skills and nuance to applying them. I don't really have an answer to that one other than still giving that work to junior staff, but they will then just utilise an LLM to help complete the job, and rightfully so, I'd have done exactly the same thing.

I think OpenAI tried doing this with some early chatgpt settings with "learning mode" or whatever that is. But an LLM just telling you what to do will never teach someone, there was something about having to go out and hunt for the answer through the sea of crap responses or wrong SO questions that really taught you something. How do you replicate that learning with an LLM and how do you apply that to different industries like Law?

For example, redlining an OEM agreement with my vendor. They pretty much all use the same template, and the objections are always the same. I tried using Claude first, and even my lawyer was impressed that it found things they missed. Then I used Claude to redline that type of contract and the other side said "your lawyer was surprisingly thorough". I had my lawyer do the final check, but saved a bunch of money doing the first three rounds with Claude instead. And my lawyer basically just said "looks good to me" by the time they reviewed it.

Or even writing a simple will or estate document.

It seems like LLMs are great for the kind of generic work you'd usually pawn off to a new associate or paralegal, but not for the important things that partners usually do.

I feel bad for the team, but will be great growth opportunities for all.

Yesterday, I had a relatively trivial bit of work in play using distance-weighting to measure stuff, and Claude did really well / but right at the end, despite having a success metric and lots of great results matching it, Claude insisted my maths was wrong and needed changing.

When i asked why, and pointed out that the suggested replacement would cancel out the distance-weighting entirely, it immediately agreed with me and asked to write its fifth project memory about it.

The other big ones it tends to mess up on without strict guardrails are coordinate systems; and measurements, often converting back and forth between a given nation grid and wgs84 latitude/longitude in ways that lose good amounts of precision; and then custom-coding haversine-equivalents to measure a distance that’s otherwise be a simple subtraction via the national grid coordinates. It’s baffling.

My example comes from maintenance in machine shops. My dad's siblings were almost all employed in machining in some fashion. One uncle was maintenance for a local large company. They decided they could move maintenance from in-house specialists and contract that work to a multi-state massive company; anyone can follow a maintenance manual right? All you need are the proper certifications and everything is cool, right? The cost was less, and the number of maintenance staff on site actually went up. Great for the owners.

But what they lost was consistency and understanding in the process that eventually led to closing the company. My uncle was there every day all year long for decades and could hear and feel in his feet when something was going wrong with a machine regardless of the maintenance schedule approved by the company that manufactured the machines. The new outside company didn't pay as well, and didn't have great benefits, which meant that they cycled through staff very regularly. No one was there consistently for more than a year or two. Even though they were technically following the manufacturers recommended maintenance schedule, machines went down. Machines then stayed down for days or weeks longer than they should've, because the expertise to repair either wasn't there, or the person leading the job would quit and they would have to cycle in a new maintenance person. They lost jobs, customers took work other places. 8 years later the company closed.

There are lessons in this kind of thing, but nobody learns them.

The problem is one of complexity, the whole being greater than the sum of its parts. Electrical engineering creates a thinking that is opposite to what you need for understanding biology or human culture.

This is relevant because LLMs have basically swallowed all written human culture. Engineering is about creating parts that interact in a clean, well-defined way, like a MOSFET transistor or a REST API. The goal is that that when you combine the parts, and they all operate within specs, you can still predict what the whole system will do.

So saying LLMs are just a step away from the Pereceptron (or an equivalent filter) is mathematically true, but almost useless. It's like saying that a modern 3D game is basically just transistors, logic, CPUs and memory, nevermind the pattern in that memory or how it came to be. You really can't ignore all the cultural inputs and feedback loops if you want any kind of useful understanding what this thing is.

It was never about cookies. Thats what the industry made you think. You don't need to show consent banners for cookies (like auth sessions). You need consent for tracking. This includes profiling. And it doesn't matter what you use, fingerprinting or tracking cookies, the consenst banner is required either way.

- Worse compatibility with the millions of Android apps already out there.

- Hardware vendors (outside Apple) all target Android.

- The usability of Android/iOS is far better than non-Android Linux phones.

- The security of non-Android Linux phones is in a bad state (e.g. lacking proper app sandboxing). Heck, even the Android emulation layers that some of these phones have, throw the strong Android sandbox model out of the window.

The answer is very obvious and it is AOSP. Even though Google is trying to close AOSP more (e.g. by not pushing source continuously to the AOSP Git repos and keeping QPR1/3 for themselves), it is still an open source system. Play Services can either be run sandboxed or emulated with microG. AOSP allows alternative systems to hit the ground running, you get great app compatibility, a tested security model, great usability, and support from hardware vendors, including vendors that are alternative friendly (used to be Pixel, now Motorola, Fairphone, etc.). If ever needed, AOSP could be forked and it would still be leagues ahead of non-Android Linux phones.

All the downsides that apply to AOSP-based systems (like things getting blocked by Play Integrity, unlockable bootloaders) also apply to non-Android Linux systems.

3blue1brown has a great explainer video on what holograms are: https://youtu.be/EmKQsSDlaa4

But there’s definitely a kernel of truth here. 1Password recently began to enforce userVerification: required[1][2], and suddenly on the desktop without TouchID I have to enter my login password to use the passkey on a good number of websites. I didn’t sign up for 1Password 15 years ago so that I have to enter both my very complex master password and very complex login password to sign into websites, the “one password” contract is suddenly broken after 15 years. This experience independently alarmed me about the additional constraints potential of the passkey standard and power wielded by the FIDO Alliance and partnering organizations addressed in TFA, which are, to be clear, useful for security in an organization. (We use userVerification: preferred now btw.)

Therefore, I believe the passkey technology is great for enforcing a standard of security in an organization and stand by my choice for my org. But as a general purpose B2C auth tech, if the additional constraints become widely adopted, user freedom will be seriously harmed.

Edit: I asked Opus 5.5 to vibe code a passkey auth solution in Go a few times without saying anything specific about userVerification. It chose required each time. So expect to see more of these. Oh and vibe coding auth without understanding it entirely is bad, please don’t do that.

[1] https://web.dev/articles/webauthn-user-verification

[2] https://www.1password.community/1password-at-home-31/user-ve...

I don't agree with the conclusion, though.

TOTP is great, flexible, and keeps the user in control; however, using a passkey in a flexible password manager (Bitwarden, keepass, 1Password, etc) really does continue to give the user control. And the user experience is dramatically easier than filling in TOTP codes.

To advocate for TOTP over passkeys, we're already relying on an application to generate those numbers (authenticator apps, password managers, etc). I see no reason to not simply continue using those pieces of software to manage passkeys, too.

For what its worth, the Total Annihilation derived project with the most attention currently is is BAR https://www.beyondallreason.info/ They decided to make it legally distinct, but it was the balanced annihilation project which was total annihilation rebalanced for better pvp. I don't play too many RTS so this is not authoritative, but BAR plays better than any other rts I have tried, They really care about the fine details of user experience there.

In fact, I consider Chris Taylor(Total Annihilation, Supreme Commander) to be one of if not the greatest RTS designers. His games don't have the fluff people appear to love(looking at you starcraft) but he gets the control design better than anyone else I have seen.

We don’t know where this technology advance will lead or settle, so it’s absurd to try to establish a working paradigm at this point. It’s like any system, the initial conditions can be extremely chaotic and impossible to model, but with time often a stable state emerges. But the stable state is impossible to identify from early initial states.

I know a lot of folks feel this, to torture other physics metaphors, sensation of jerk - acceleration of acceleration. It’s an unpleasant and dislocating sensation. A world that felt safe and stable suddenly isn’t, and not in a micro tragedy sense but in a global realignment sense. This happened to factory workers who had enjoyed generations of stable work, farmers more slowly and just as surely.

This is what the late stages of scarcity feels like. Labor of various types devalues rapidly. Our exchange of meal and health coupons for toil cracks, and people realize their labor wasn’t godly as great books told us, but simply needed for want of an alternative. The realization that our labors might not be valued any more, and that our sense of purpose is shaken, coupled with the fact we’ve tied bare survival to our toil in our labor, is mortally tightening. No wonder people are grieving publicly.

But maybe our purpose isn’t to toil? Maybe we’ve passed peak population, and as toil is less valuable, we need less people and that’s why population is declining. Maybe we don’t need to exchange food and health coupons for toil, maybe mathematicians don’t need to rationalize their value to pursue mathematics. Maybe they can pursue it because they can’t help but pursue it, and our ever improving automations can produce their meal and health coupons?

But it might require Dr Tao to take an AI generated cancer medicine some day.

Russia went all in on "digital sovereignty." This is partly because the Russian government cares deeply about censoring what their citizens can see (which they can't do if those citizens consume content from American services), partly because the American services themselves cut Russia off because of sanctions-related concerns.

This makes Russian datacenters a much more attractive target for the Ukrainians. While Russia can (and does) bomb Ukrainian datacenters, the fact that Ukrainians are so reliant on US infrastructure (and could rapidly move local workloads to western clouds if the need was great enough) makes such operations far less attractive. It's much easier for Ukraine to strike some facility near Moscow than it is for Russia to strike Ashburn, Virginia.

Given the geopolitical concerns from both the east and the west, this makes the question of whether Europe should move content off of US infrastructure a lot more complicated.

Maybe this would remind ordinary russians that killing innocent civilians every day in Kyiv is not good

Great story, programming and mechanics.

Where I live, in the European Union, the price of whey protein concentrate had doubled around the beginning of 2026 and now it has reached even a triple price in comparison with a year earlier.

For a few years in the past, before the price increase, whey protein concentrate had been cheaper than the cheapest meat, per its content of protein. Now it has become about 50% more expensive than chicken breast, per protein content.

This is annoying for me, because where I live there were only 2 ways for achieving a diet satisfying simultaneously 3 criteria without including meat: sufficient daily protein intake, not exceeding the calories threshold that would cause rapid gain of weight and price not greater than an equivalent diet with meat (there are a lot of plant protein concentrates, but all are many times more expensive than cheap meat, at least here).

One way is to extract at home gluten from wheat flour, which is an activity that wastes too much time and too much water, the other way was to use whey protein concentrate as an ingredient for various dishes made from vegetables, to increase their protein content.

At the current price, the use of whey protein concentrate is no longer attractive, despite the fact that whey protein is the ideal source of protein, with optimum amino-acid composition and perfect digestibility.

The FT article believes that this sudden increase in the demand for whey protein concentrate was caused by the widespread use of drugs like Ozempic for weight loss, which also causes loss of muscular mass, leading to the use of protein supplements to prevent it, and also to the new US dietary guidelines, which recommend an increased consumption of protein (the old US recommendations for daily protein intake were notorious as much too low, i.e. they were at a level that prevents death, without ensuring good health, but the new recommendations might have jumped on the other side of an optimum intake).

The only possible positive consequence of this huge price increase is that it might convince some people to invest in the production of whey protein concentrate from cultures of genetically-modified fungi, instead of from cow milk. There already exists a strain of the fungus Trichoderma, whose genome was modified to secrete cow whey protein, instead of an enzyme secreted by the wild fungi.

Unlike with "lab-grown meat", there is no doubt that it will be possible to produce protein powders of whey protein or of egg-white protein in a cheaper way from fungal cultures than from milk or eggs.

Whereas you take great care of your local community by even refusing to live around anyone not as well off as you ?

Imagine the horror of having to live within 10 miles of a low class smelly poor plebeian, you might have to lay your eyes on their 3 days overgrown lawn from time to time.

In the UK, 'stamp duty' is something that buyers pay (as opposed to sellers), but sellers also have to buy, so everyone pays to move and everyone stays put to avoid paying.

(And of course no one builds, which is the biggest issue).

I'd say it's OK to defer your property taxes until you sell (or die), but it shouldn't be the default, only something you apply for if you're in a vulnerable position.

I think it's a great idea to let land taxes create the incentives to build efficiently.

Lobbying is indeed corruption with a suit on.

We are not talking about a properly configured Windows workstation. We are talking about corporate PCs configured by an IT department that is completely oblivious to the needs of developers, most likely because developers are not their bosses. WSL is effectively an escape hatch.

Usually the stated reason is "security", and developers tend to do the greatest sin of corporate security: they run arbitrary code, and they even have the nerve to call it "their job"!

I don't work in corp IT, but I am convinced that people there are not stupid and they are fully aware of the "shadow IT" aspect of WSL, but they give it a pass because they are also aware that people have "real work" to do that would be impossible otherwise, given company policies.

When I was a junior developer I did see programming as an art but I really don’t see it anymore. What is important is the output. And does the output is in accordance to its design and functions as intended.

You can produce great software with any code or programming languages that you can think of. Some tools and practices produce them faster and/or with higher quality and correctness.

You can replicate the same product with almost any programming language now and good software practices now. AI makes the writing process a breeze but doesn’t solve what is art: the design and the vision (why and what). Therefore I think you can’t call the process of production an art. It’s really not that special.

A bad manager takes personal credit when things go well and finds a scapegoat when things go bad. A good manager gives credit when things go well and takes responsibility when things go bad.

It's wrong to scapegoat the agent. You, the human who executed the agent, made the decision to execute it. You gave it credentials, and set up insufficient guardrails. Don't scapegoat the agent for your poor decisions.

Likewise, it's good to have humility when you get great results out of agents. Yes, the agent made the good results happen. It's OK to tell people that your choice of agent is amazing and did an incredible job. In any half-decent work culture, that should rub off on you as well.

Decision laundering is a great term. There is ultimately a human who gave the program free rein. You can't say "sorry my car hit you, it's not my fault" especially when the person driving works for the car manufacturer too.

It really depends on your perspective of what the point of the hobby is. Were decomps/recomps a hobby because people wanted to make sure their old games could run on modern hardware and be extended with mods etc., or were they a hobby because people enjoyed the actual process of using Ghidra and reverse engineering by hand, and the decompiled game was just a side effect?

Personally I think it's awesome that PC98 Touhou games are getting recomps. Being able to play the series natively on modern hardware, and all hardware for the foreseeable future, is great and preserves the legacy of the series. I understand that for anyone who enjoys the actual act of reverse engineering it feels like cheating, though.

I think people will get over it. Music is a good example. Drum VSTs and piano VSTs are going to sound way better than you unless you're a 99th percentile musician playing on a 99th percentile physical instrument. But people still play the piano and drums, because it's fun. Same idea; the existence of AI doesn't stop people who love the act of reverse engineering from doing it.

* Illegal to collect every plate for later analysis

* Must delete "non hit" plate images within 3 minutes

* Strict forbidden to upload "non hit" camera imagery off the device

I would add that it should also require a warrant to access the ALPR data, but it's a great start.

https://gc.nh.gov/rsa/html/XXI/261/261-75-b.htm?utm_source=c...

Counterpoint: there are a lot of one-hit wonders, and they vastly outnumber the idea-factory people. This is not to minimize those people, a single idea can be very successful (see Zuckerberg), but it doesn't mean your subsequent ideas will also be great (see Zuckerberg)

Kind of funny how much hype they can all get out of this, but Qwen really is the little engine that could. Great to see open weights (if not open source) driving the whole ecosystem like this though.

I feel like it's a sign of what's to come - large AI companies with resources academia never had, at some point deciding that selling access to AI is not as important as just doing the research themselves with models inaccessible to anyone else. The end game is all scientific advancement will be done by one or two companies, who will reap all the profits. Why would they give access to a model that develops a cure for cancer when they could develop the cure themselves and get all the money?

(Kind of like Amazon coming out with their own products - use the greater community for ideas, then become a competitor at a scale they often cannot match).

I got another great offer after just 1 interview that I took, so I never went through their process, but it looks very exhausting to me. Being rejected after investing so much time must also feel awful.

Oxide continues to be one of the most inspiring companies in the space, I was just encouraging someone to apply there yesterday. :)

Nothing wrong with that.

One was even a non-profit, which should be more concerned with the well-being of humanity (which they assure is in great danger from what they produce) than the continuation of the company.

- NASA (with the TESS mission): https://www.zooniverse.org/projects/nora-dot-eisner/planet-h...

- ESA (with the CHEOPS mission): https://www.zooniverse.org/projects/bmerin/planet-hunters-ch... (Very new! Will be announced next week!)

If you are logged in with an account they will even recognise you and add your name to the paper!

So instead of playing alone, join the common effort, this is how science will progress efficiently!

I do have to say, however, that BSN syntax isn't good and is getting worse. There are too many sigils, and -- to separate list elements (!) is an indication that it's been designed into a corner. The fact that it's not LR(1) should have been an indication that it was misdesigned. The scene format should be redesigned to be editor-first, with ease of VCS merging as a paramount consideration.

The finished product benefited greatly. There's nothing quite like working with a genuine expert who knows the rules and tradeoffs on things like competition, color, etc. Author did good on providing the illustrator with a detailed brief. When the illustrator takes those guidelines and applies their style and training to the product, magic happens.

The J1s are then packed like sardines into houses where they pay rent by the week. Why by the week? To remind them they're expendable and can be deported at a moment's notice by the restaurant/hotel/amusement park/etc. they work at. The worst I've heard is 16 people living in a 3-bedroom, 1 bathroom house.

The J1s themselves are typically great people. A lot of them do everything they can to stay in the area via Green Cards or whatever. Business that treat their J1s like shit are usually blacklisted by locals, too. We'll actively tell tourists not to give those businesses any business and why.

It makes me wonder why there aren't more American young adults taking these jobs. I think it would be great fun to live at the beach for the summer. Maybe the wages are too low, or American kids are focused on internships? Or maybe the European workers are better employees? I don't know.

I just setup Windows speech to text for him last week and it's great to see how he can write an entire page in 10 minutes, it would take him days using the keyboard.

But every single sound he makes with his mouth ends up on the page too.

Those turbines aren't great for longer periods, they're pointlessly inefficient if you don't need their ability to peak for short periods.

People say that agentic development is great because you can churn out so much so fast. But that doesn't mean that any of it will be truly good and reliable.

The things that are truly insightful and solid end up being used exponentially more, which makes the linear cost of extra development time (asymptotically) insignificant in the cost/benefit equation.

When you have the mind of "the models are wrong", you tend to be brought there by select counter examples, and in that space you only feed on counter examples. "The experts are wrong (and you can be right in 15 minutes if you consume this media!)" is abundant.

The actual reality is "the models work great until they don't". Not "the models don't work."

Ironically, the experts, the ones building the "wrong" models, tend to be the ones most aware of this.

The amount of energy and material wastage in developed countries is enormous and abominable. Large screens being on 24x7 without anyone watching, cities explicitly designed for forcing the use of vehicle, consumer life-style requiring huge supply chains, powerful cars carrying just one person, ... and so on.

Business drivers will continue to drive the greed of people and nations and act against any measures to fix the problem. A business knows nothing about climate or human good.

This project looks great, well done, BUT it seems very odd to be "grumbling" about Pi when you're explicitly targeting a separate audience (GUI users).

There are other GUI tools like Juggler that are quite popular, like cmux, that it seems better placed as an alternative to.

It sounds like you're advocating for arXiv to become viXra. On https://vixra.org/all/2610 the second paper is currently "Correct Interpretation of the Great Discoveries in Particle Physics: I. Reconsidering the Higgs Boson via Vedic Vortex Structure", followed by a paper arguing that classical electrodynamics is bogus, "Gauss’s Flux Theorem Does not Hold in Time-Varying Electric Fields", and then "On the Boundary Problem in the Origins of Matter, Life, and Consciousness".

I don't think arXiv will continue to receive "quite some funding" if this is what it contains.