epolanski

Antirez spends weeks planning with multiple models till he has a clear architecture and understood the constraints before he writes a line of code.

I was a chemistry researcher this is true in all science.

I'm not sure LLMs can transform this, the incentive is to get more results in your nich, jumping topics don't help unless you have genuine interest and reason to.

People are insecure about their leetcode black belts and react slop not giving them cushy jobs anymore so keep missing the forest for the tree.

They lack the empathy to understand where and why you're struggling.

I've given private math lessons and seen students struggle with ai, even though ai gave the right answers.

My intuition is that humans spot xy problems easier when teaching (user ask x but really needs y), whereas llms will oblige writing about x.

Mathematicians emphasize definitions not labels/names.

It doesn't matter if natural numbers include 0 or not, what matters is how you define them, not how you call them.

This makes them also bad at naming things because...there's a definition anyway.

Most other fields do not have or can't have the same luxury, so naming might be more thoughtful.

I wouldn't want a GPU datacenter close to my house either.

They should buy land in middle of nowhere, many miles from any closest town, no less than 5. And they should build and provide their own electricity.

I don't see any other solution to make everyone coexist in peace.

Imagine it was your house a mile from such a monster. Eyesore, noisy.

Another issue is that they bring very low to no jobs to locals besides some initial construction work.

I think that would also be a bad idea, as all models opus 4.6 got increasingly smarter, but also crappier at following instructions or genuinely assisting.

They just try to figure out what the goal is and hyper focus on solving it.

Heh, even just telling fable don't commit doesn't work half the times, let alone more complex instructions.

If it was genuinely useful, we would've long reached the point where you train a model on a previous one's output in an ever improving loop.

But this doesn't actually work.

This is BS to pressure politicians.

Even an openai's guy (head of something made up) called bs on the idea you can train something like k3 by distillation.

Anybody I know who works in LLM research says that distillation is either useless or merely useful in post training to show "correct" behavior.

And even then you don't get a competing model, if RL on good prompts was that useful, all labs would've long skyrocketed in capabilities just by looping on increasingly better prompts, yet that doesn't work.

I respect your opinion, but behind winners imho there's 99% being in the right place at the right time. And people change and so do their motivation.

I'm not saying top talent doesn't make the difference, I'm just saying that CVs and impressions aren't always completely related.

Only the Russian business is in hands of Russian investors since 2024.

And again, you can check that Kagi answers any query about Russia yourself.

1. Yandex is dutch controlled, the russia-related business got sold in 2024 to Russian investors.

2. Yandex was never owned by the Russian government, albeit government-related censorship happened (but so does on Twitter/Google/Meta etc as they all have to comply with local laws where they operate).

3. Kagi answers questions about Russian war crimes. You can try yourself [1]

4. There's no such thing as a legal war, the closest is a UN sanctioned peacekeeping operation.

[1] https://kagi.com/search?q=what+war+crimes+did+russia+commit+...

This isn't electric, it's a balance bike for kids you push it with your feet.

In any case there's endless electric bikes on shops.

It's between odd and creepy, not sure where the cool is.

Even if hardware capacity increases, it seems clear it uses way more tokens, so I don't expect parity with other competitors on that front.

On the other hand I expect K3 future refinements to be massive and more efficient.

Clearly, for you.

I've read the same opinions about Opus and yet it was gpt 5.5 pro via api tackling the hardest problems.

I have used now k3 for 3 days and it has consistently tackled difficult problems sol max could not (orientation optimization algorithms of random 2d shapes on a rectangle for glass cutting).

I have also other beefs with Anthropic models which have been getting smarter and more capable since 4.6, but increasingly worse at acting as assistants, they just want to "do" stuff their own way and ignore instructions often (even simple ones like not to commit, let alone complex ones).

I don't see a difference in capabilities between k3 and fable, but k3 is slow and expensive. Burned through my monthly plan in 3 days.

I don't think I write or read or code anymore, bar prs from juniors or senior colleagues wanting a review.

I didn't think it would work just 6 months ago, but reality is that at this point AI writes better code than me and I'm not the average developer, but someone who loved the craft and was good at it.

Lots of effort was required to get the repositories to a good level, best practices, documentation, etc, but reality is that once you do that and have strong rails most of your work is having it to write a plan focused on business logic, review it, have it derive an implementation plan, review it and then it's mostly on its own.

Codebases have never been healthier, cleaner, better documented, consisted and thoroughly tested as they are now. There was just no spare time and mental energy to bring them there before, now there is and experimenting to get there was cheap.

Needless to say I no longer enjoy the job anymore and thinking of changing domain. I loved tinkering about implementation details, etc, but the job nowadays is more of qa and architectural design than writing or reviewing code.

Google somehow managed to snatch defeat from the jaws of success with their AI products.

HN lives in a bubble.

I have German/Italian/Polish clients virtually all use Gemini and NotebookLM. Talking insurance, banking, consulting, legal.

The real world doesn't look at pointless benchmarks on writing react tailwind crap, they are already google suite users, get the tools, test them and adopt them, end of story.

It's going to be like with angular, never mentioned on the net, widely used in the real world.

It's more complex than that, especially as post training is often goal based.

I have a similar use case at work for previewing construction material and such in our catalogue applied to user uploaded images.

Results are mixed, expensive, but it really feels you're few months off the next improvement to really nail it. It's already good enough.

Wonder what Qwen image will provide over nano banana.

Every person I know me included got lost there multiple times, yet some users here get offended by this somehow.

If you take the gross area of Shinjuku, including its multiple maze-like levels, it is comparable to the size of city centers of Florence or Bologna.

Not only that but Shinjuku is under massive renovation work which makes the signs genuinely outdated at times.

I've been in Shinjuku few times, I've always got lost.

Found a "hack" where I would go upstairs to the mall and then go down closer to the spot I need.

Shinjuku is way too massive (and now has plenty of diversions due to work om the station itself) to be understood easily by a tourist.

To put it in perspective, if you wanted to say "hi" to every person entering Shinjuku in a single day, and it took you one second to do so, it would take you 35 days of uninterrupted greeting.

At this point it's just a matter of having enough ram in your consumer computer.

Until we reach a terabyte of ram at affordable prices imho this isn't going to happen.

I never saw an ad except when opening Xiaomi own apps which I don't use.

I don't share the definition of over engineering.

As engineering is the act of solving technical problems, over engineering is about putting too much engineering effort on aspects, features or products that don't have a linear payoff to the budget spent.

E.g. I worked in a company that was obsessed with unit test coverage metrics and the effort of maintaining the test suite was considerably biting in the ability to move the product. The ratio was 25% of product and 75% of unit tests.

And the payoff was small if not even negative, impacted morale, productivity and actively pushed back against refactors, because any large refactor was met with a disproportionate effort in unit test rewriting. Let alone the fact that as you were mocking external dependencies, and the mocks costed engineering effort and internals digging, it also suppressed work to keep it up to date.

I never got to convince the org to shift focus on E2E testing, which answered the real questions: does the product make money and works as expected. Uni test had to be used when writing a parser, not to validate some react scroll component as browser apis and the triggers mimicking costed days of work.

In fact we often had gigantic all green test suites for broken products.

That's to me over engineering an aspect of the engineering cycle.

This makes everything worse, not better.