snek_case

Probably AVX-512. AFAIK Apple CPUs are still using ARM NEON instructions with 128-wide SIMD registers. I guess they are banking on you using the GPU if you want to parallelize those kinds of workload.

It might be conceptually similar to a single-output-token LLM (sort of). LLMs output next-token probabilities. You can ask LLMs to output yes/no, or to output only a color, or only a digit or something like that.

In this case I would imagine that they probably embed your input data into a vector space, and they embed your questions/outputs into another space, and manage to predict probabilities/classes/scores for your outputs very quickly. Embedding the output classes/questions into a vector spaces gives you something you can reuse across runs cheaply, as opposed to an LLM where you can prefill the KV cache but this is an expensive operation in terms of memory.

Agreeing with you. Trump's attitude on the global stage has been to disrespect, insult and threaten every ally. Come at everyone with a "we're better than you" attitude and show that he can't be trusted. The only rational response is to take steps to be more independent and find ways to work around the US. What's the alternative? Give in to a narcissistic bully who won't keep his word and will inevitably come asking for more later?

In the medium to long term, Trump is very old, his approval ratings are in the toilet. I hope that the democrats do well in the midterms and put up a viable candidate to beat him next election (if he's still alive). I also hope that the republicans manage to put up someone smarter and less morally reprehensible next time. Because I mean, this guy doesn't actually represent any kind of conservative ideal in any way. The only people he appeals to are those one or two standard deviations on the left of the bell curve.

If the next US president is halfway sane, Canada and the US will likely begin to repair trade relationships, but I think that Canada and the rest of the world will be a bit more careful next time. The US may never regain the international standing they enjoyed before. Trump has made China look like the more stable and rational trading partner, and accelerated the decline of the US.

How far can you get with Qwen 3.6 35B A3B? What scope of task is it able to do?

Ad impressions typically cost money, so there's a case to be made that this is sort of self-correcting too.

That being said, IMO no website should be able to freeze your machine. This is a bug. Steps should be taken to fix it.

Most people (and most companies) can't tell bad ideas from good ideas. This is also true of LLMs to some degree. They need post-training in specific domains (e.g. coding) to become competent in that specific area.

You can use it for things like classifying documents, spam detection and detecting certain events, etc. You're right that it's not strong enough for coding but it can still be quite good for question answering and such. Or even simple things like converting documents from one format to another.

And yeah, for NPCs, this model and their other A1B model can fit inside of a game and run at a decent speed on CPU-only. It would probably be workable for game characters.

I would like to see an examples directory with a diverse set of example programs. What would be most interesting to me is programs with non-trivial memory usage and data structures, because that's where differences in the memory model would matter. I want to know what's possible and what's not allowed in this language. What makes this different and better than Rust's borrow checker.

The weights are very unlikely to be on the chip itself. That wouldn't work for SOTA models that are terabyte scale, even quantized. This is probably an accelerator for specific kernels in the model, but the weights are likely loaded from memory. The chip may have SRAM to store some of the weights temporarily during inference.

In order to cure cancer with AI, you would need to give it complete control over the experimental pipeline, including probably direct robotic access to a web lab, which comes with its own set of risks. It's just not the same problem space as coding at all.

I think that people will argue around that. Self-driving cars are already strong enough to kill people and even if you only made the robot strong enough to lift 15 pounds of weight, that's probably strong enough to slit someone's throat with a kitchen knife while they sleep... There probably isn't a level of strength that's weak enough to be completely safe but strong enough to be useful.

So at some point you have to trust that the tech is safe. Both in terms of "robot won't go off the rails", but also in terms of hostile actors can't remotely take over your robot while you sleep.

In terms of sleeping, personally, I would like for the law to mandate that robots must have a physical off switch, in a very visible location, that physically disconnects power. The switch should be illuminated while in the ON position. What makes me a bit pessimistic there is that we don't even have laws to mandate webcam indicator lights (e.g. a very tiny red LED) must be ON in hardware.

The thing with drug discovery is that the bottleneck is not coming up with new molecules, etc. It's the multi-year, extremely expensive clinical trials that often don't even succeed. I doubt it will be a big moneymaker for Anthropic and OpenAI, because the best they can hope for is to sell a software product to pharma companies, but the pharma companies already have lots of scientists with a million promising molecules already in study they will never even make it to clinical trials.

In other words, can AI tech help? Probably, but I doubt it's going to be a major money maker, and it won't make a ton of money overnight. It's not like Claude Code where every programmer can begin using it instantly.

Probably because the team lead is a web developer, and whatever he says goes. Companies are typically very hierarchical, even if they like to claim otherwise.

People widely report very high memory and CPU usage.

Not all of the memory usage is the fault of Bun/JS. Some of it is likely memory leaks (holding on to data it shouldn't). However, some of the blame does go to using JS. In Rust you can tightly pack your data if you want to, but in JS you can't have nested structs for example. Every object referencing another object is two separate chunks of memory, each with a header, and a pointer from one object to the other. GCs also use more memory. The overhead adds up.

Building a CLI program in JS is a bad choice, and nobody should be defending this decision. Especially since Claude Code is very much able to write Rust code, and they've shown it can port code. Just port Claude Code to Rust directly.

The reason it makes people uncomfortable is because people have been using it to alter images of real people, and they've done that in a public place (twitter/X), for everyone to see. So it gets in the deepfake realm, which is illegal in many places.

Elon's public take so far has implied that he wants Grok to have better ability to reason about math and physics, thinking that this will make the model more rational (and so less biased). It's possible that they have internal RL post-training designed specifically for that. It's clear that whatever they've done hasn't made Grok align with Elon's beliefs though. Not sure if that will last or if Elon will eventually push to make the model align to his own political beliefs.

Yeah I mean, if you don't know how to code, you just know how to prompt, you have no idea how to tell what's a good solution vs what isn't. The best you can do is have the model figure it out for you. You also have no idea how to design a good API, or how to break up a system into modules, etc.

The issue is probably that many managers can't really tell the difference between a good programmer and a vibe-coder. The vibe coder ships a lot of PRs. Maybe they themselves ship some vibe-coded PRs. They hate the idea that programmers might know better than them.

It looks weirdly petty that they are going after her. Everyone knows that Meta is a shitty place to work. There have already been countless articles. Their reputation as an employer is almost Amazon-tier. Is there much point in trying to silence one person?

Simple software is also easier to optimize. The simpler it is, the less weird corner cases you have to deal with.

I've come to understand, talking to a friend who is also in the startup world, that there are some CEOs who run a "portfolio" of startups. It's pretty weird. I don't really understand why investors would put money in a company that only has a part-time CEO.

I was working at an AI startup, and I saw our CTO lie in a demo to potential customers. I know that startups sometimes have a "fake it till you make it" mindset, but the guy straight up used a product from another company, presented it as our company's product, and faked numbers. I saw him completely misrepresent the capabilities of our product several times. Unethical and most likely illegal, I felt super disappointed, but I didn't immediately quit.

I quit later, as it became increasingly clear to me that this guy knew nothing about technology, didn't care, but also had a fragile ego where he had to present himself to the company as being in charge, even though he was the worst person for the role. To top it all off, it also dawned on me over time that we basically had an absentee CEO who was working only ~15 hours a week at most. Then when I quit I found out there was a third co-founder who owned a huge stake of the company and I did not even know existed while I worked there.

When I first interviewed, the CTO seemed like a nice and friendly guy, I didn't immediately see red flags. This was my first startup experience. I'll try to research things better if I decide to join one again. I might also just not join unless I can myself be a co-founder. Fuck reporting to incompetent twats.

Currently taking a sabbatical. I decided to take the summer off. I'm working on personal projects. Lucky enough to have good savings from a previous job so I can afford to do this. I'm planning to take gradual steps towards returning to work near the end of the summer.

To get tenure in STEM, you need to publish 1 to 3 papers a year. If you only publish once per year, you'd be on the low end, so you always have to think about your next paper. You always have to try to work on something that would be publishable within the next 6 months to a year, but preferably within 6 months.

I agree there's a path where this encourages people to be even more sedentary and lazy. On the other hand, if I cook every meal and clean up after, plus take care of home cleaning, I could easily spend 3 hours a day on a variety of home tasks. A robot could potentially prepare better quality meals from fresh ingredients and save me 3 hours a day. It could also fix holes in my clothes and do other tasks I'm just not motivated enough to do. So the way to think about it is like you just gained a huge amount of energy and free time to do things you weren't doing before.

I think people are also literally fighting datacenters. As others have said the increase in energy costs is a problem for the average person. Not only is AI potentially competing for your job, it's also competing for your access to energy to power your home or your vehicle. Energy costs also affect the price you pay for basically every good and service.

Then there's the fact that many of those datacenter are being built over what would otherwise be usable farmland. I'm sure many will say "it's not that much land", but then tech billionaires would like to build datacenters the size of Manhattan. What for? To train a bigger LLM? Yay?

It's a win for consumers as well if you can get better video quality or more reliable calls on a slower connection.

Maybe synergistic with tokenmaxxing. You should be burning more tokens, and you should also be making more trades.

The people who make that comparison are just showing how clueless they are IMO.

I wish the tech field wasn't so full of clueless grifters. The most frustrating thing is that this kind of people, the kind who loudly and confidently assert bullshit claim based on insufficient knowledge, have a knack for positioning themselves in positions of power.

The difference is that LLMs are not compilers. You can't trust the output to be correct. They routinely make bad design choices. If you're prototyping some kind of throwaway MVP, you're just sketching something, it's probably fine not to review it. If you're trying to build a piece of software that's going to survive for years, why are you doing this? The tech is clearly not yet fully mature.

Just yesterday, I was trying to use Claude Opus 4.7 to debug an issue in a program that I wrote, and its solution was to remove a feature, change the design to eliminate the problem without consulting me. I only found out that it had removed this feature through testing. Imagine not reviewing things like that. How can people argue for this with a straight face? We'll probably get there eventually, but there's no need to rush before the tech is ready. Doing that is just being clueless.

Yeah this is pretty shady. The S&P 500 in particular has fairly strict criteria (e.g. 4 consecutive quarters of profitability) and those criteria exist for a reason. They made me more comfortable buying the S&P 500 knowing I'm not buying pre-revenue companies. This is a bad precedent to set just to please Elon.