gobdovan
gobdovan on Hacker News
Even without an explicit DB it would have been quite strong, core Stockfish would be enough to beat any human as well. Ofc learnings from strange games would still be compressed somewhere in the eval functions.
Italy has a pretty rich history of controversial PMs. Pre-WW2 Crispi, literally Musolini, recents Berlusconi, Andreotti, Craxi. Don't think I know any other foreign country by how controversial so many of their leaders were.
Did Italy ever had a PM that was not controversial?
I'd say that's about right. The way people usually code in Python, mro is quite stable and most lookups would be cache hits. You can still get more weird invalidation in Python cuz they let you change stuff like `__bases__`, or you get some weird `__getattribute__` change in some place that would force you to invalidate more than you'd initially think should be necessary.
Hegel might see the demand of justification from AI as a form of alienation that could make us more deliberate in our choices, particularly in a time where we give up more control away without examining it properly. Unconstrained optionality is not the same as freedom and sometimes putting up your own constrains can increase rather than diminish self-determination, especially when you're aware of the barriers.
Yes, it does beat -O3 C sometimes. I took code from diverse 'language speed' benchmarks from the internet and had to rearrange them to get -O3 to beat equivalent-looking JS (Node 25.8.1 w\ V8 14.1 and clang 21 on MBP3).
> What was that code?
literally:
```
function numericKernel(a, b, out, n) {
for (let i = 0; i < n; i += 4) {
out[i] = a[i] * b[i] + 0.25;
out[i + 1] = a[i + 1] * b[i + 1] + 0.25;
out[i + 2] = a[i + 2] * b[i + 2] + 0.25;
out[i + 3] = a[i + 3] * b[i + 3] + 0.25;
}
}
```
Have you considered exposing these optimisations through a library so programmers can request optimisation passes and have the compiler check the preconditions?
Asking cuz I'm preparing an article on how fast JS can get and found the same thing, you can get in 10% of optimised C-code with basically any code if you torture V8 enough (excepting SIMD where there seems to be a bug in V8 where it doesn't optimise to SIMD even when beyond obvious). Figured out I could extract it in a lib and do it more streamlined but not sure how well manual transforms generalise.
Also, when working on the roc-deflate, did you find recurring transformations that could fit that model or were most improvements specific to compression?
Depends really. I tried some experiments towards a pretty uniform relational language for expressing ordinary logic, temporal logic, requirements etc. It does work, but there's some unexpected obstacles.
You'd think performance would be easy to achieve because you'd just write clear, high-level declarative facts and LLM + good optimiser + proof certification system will get you there in terms of performance. But you'll be surprised that relational/declarative specification gives you so much behaviour 'for free' that choosing what to actually compute from those relations is a very deliberate choice. It starts to look like finding an imperative-like path of demand through the relations. It seems that you get exactly the same problems but one level up.
Does the thought experiment help us to identify a solution for these problems?
Well, everyone being tabula rasa has a point tho. Technically there's no unique physical cause that make development unfold the same for all organisms, or even all humans (see Waddington's epigenetic landscape). It just happens that we develop along very similar trajectories and can draw equivalences between how I think vs how you think. So, from an ontological perspective, if you squint a bit, we all are tabula rasa. But from a phenotypic perspective, of course it's bonkers, you can't teach a frog linear algebra.
In before: Goomba fallacy. But I can't notice how there's claims that: LLMs are nondeterministic and cannot be relied upon to produce the same thing AND they produce the same texts, same UIs, in the same styles. It frustrates me too, but what a solution for both problems would look like?
Also, if all LLMs would write in the style from the article, in how much time that would be the recognised LLM style people would get sick of? I suppose longer? But even the current style of LLMs was palatable when they first came out.
I'm not defining education as passing tests. A test is just a measuring instrument.
In fact, your claim is strong and quite easy to test. If watching the series gives people a better fundamental understanding which makes subsequent learning more efficient, take matched novices, randomise them to watch the series or not, then teach both groups some later lin alg materials and measure learning gains on novel problems.
That's the kind of test I mean, not 'can you grind through the course exam'.
If your claimed deeper understanding neither improves what someone can do afterwards nor how efficiently they learn related material, I'm not sure what observable content the claim has.
It's much more likely that the students who take the time to watch supplementary materials are the ones who'd do better regardless of the quality of the given material. If you have access to fresh learners, I'd say it's a good opportunity to test how good those series are.
I'm actually quite optimistic, was pointing out that, at least to to me, AI slop already produces better results than big educational content creators. For something concrete, even if not classic math, I credit direct LLM interaction highly for accelerating my understanding of CPU design and compiler tech.
Even if you find them intuitive, they produce no learning gains for anyone. Take a lin alg test, watch the 'Essence of lin alg' videos, take comparable test again. I guarantee you won't see any change even if you 'understood matrices better'. They are pure mathporn.
I tried watching 3b1b over the years, really wanted to enjoy it. But I never found the way he presents things intuitive and never got a palpable intuition that helped me with math classes or work.
I saw some clearly AI slop made videos and got much better understanding of topics, since they don't do foreshadowing for an awe-inspiring connection that we'd admire together after 15 min videos, the clankers just get to the point. Not claiming those videos are too useful either, but I did however see accelerated learning gains from asking LLMs to explain (and do Bret Victor-style small interactive demos) math and algos, something I have never experienced from any of the educational STEM Youtube channels.
> These projects exist, but aren't getting amplified.
Name them
There's already Youtube games (youtube.com/playables), where all games are hyper casual ad-driven low-effort games published by companies like Voodoo or SayGames. I believe Playground would compete with those more than with regular PC games. There's no way slop from regular people could be worse than what's already out there for mobile.
There's multiple reasons people play games. Some people are competitive and play a game thousands of hours to get better at it. So there's always a market for longer lasting games.
Yep. Simplest is browser game with three.js, the only lib you need. But for low-poly 3D, you can basically tell the frontier LLMs to build yourself the rasteriser itself in whatever language you want (not recommending it, just saying they're that good). Literally just put your comment into Claude Code or Codex and see it do it.
They handicap our little companies more than the incumbents. The Commission itself acknowledged that regulatory costs affect SMEs proportionally more than bigger competitors [0] and the OECD explicitly found that policy failures and regulatory barriers tend to hurt startups more than incumbents [1][2]. The way I see the broader pattern is: more regulation -> larger bureaucracy -> larger bureaucracy has more scope and more incentive to regulate -> incumbent absorb growing compliance burden easily -> entrants get hit harder + ad-hoc regulations to hurt established outsiders even harder.
[0] TOOL #22. THE "SME TEST": https://commission.europa.eu/document/download/5d2011d7-5470...
[1] No Country for Young Firms? https://www.oecd.org/en/publications/no-country-for-young-fi...
[2] OECD, about Germany 2025, e.g.: 'Industrial and labour market policies have tended to favour incumbents and hampered the reallocation of resources to young and innovative
firms, weighing on allocative efficiency and slowing down structural change' (https://www.oecd.org/content/dam/oecd/en/publications/report...)
If AI would be that cheap and good, why not open a company and leverage it for yourself?
> If I were OpenAI or Anthropic I’d be racing to make their products as sticky as possible bc ppl will flock to what’s cheapest otherwise.
Hopefully people will flock to whatever product is making its mission to be commodity and the easiest to replace. Really don't want another free ingress, 100$/TB egress Cloud situation.
Well, programming feels that it will get closer Smalltalk. For webdev you get pretty mature HMR, you look at the interface and sometimes just point at things and tell a model to modify it. So pretty close to live medium +short feedback loops.
If the big labs wouldn't have the incentive of hiding the code and making you rely solely on the models, you'd probably also see more inspection tools and contextual changes when the model does them.
In EU, it's politician > industrialist >>> EU citizen > outsiders.
It's almost never only about consumers. Via tech regulations, they protect European incumbents first in effect. They see US is ahead and make laws to destroy their moats. If Apple was in EU, you wouldn't have saw universal usb-c, because it would have hurt an EU company. But the more you look at how tedious is for a non-EU company to sell to EU customers, the exemptions they don't get, the specialists fees they got to put on the table, you'll see EU is more coherently described as protectionist than pro-consumer.
That's now how it works. You generally can't know if a statement is provable/unprovable under A. In most cases, there's nothing to identify a new corpus.
Not sure why people are stopping at 'AI can do advanced math!' and calling that the field is gonna transform due to LLMs specifically. Pretty much any meta-formulation of math gives you some initial rules and a goal. From natural deduction and sequent calculus to dependent type theory and whatever flavour of category theory you choose, at a sufficiently abstract level, you'll end up with some form of search problem. ATP folks foresaw machines doing automatic proofing for decades, but it seems that noone had enough incentive in pursuing it at the cost of suffering the tediousness of the encoding process. Now AI labs took the opportunity to do something impressive and have shown that it's possible to close the ergonomics gap. But the deeper achievement is the decades of ATP and proof-assistant work that make machine-generated proofs mechanically certifiable in the first place.
I am pretty sure in the next couple of years we'll see more evolved sledgehammers [0], maybe with LLM-like tech playing a role in encoding theorems, that will allow us to prove deeper than we can do manually, then help with the archeology and extraction of learnings from observing the 'trace' of certified proofs. So instead of explore -> understand -> solve, we'll get solve -> compress -> understand.
[0] e.g. https://isabelle.in.tum.de/website-Isabelle2025/dist/Isabell...