Sabitlenmiş Tweet
is this a conspiracy? is my guru @yacineMTB telling me the truth?
ily⚡️@0xIlyy
@squarecapo @0xluffy we are related
English
capo
5.4K posts
capo
@squarecapo
CLAMP DOWN. transpose. dominate the fretboard.
Katılım Aralık 2024
527 Takip Edilen338 Takipçiler
functions are deterministic distributions, so starting from prediction is starting from the general case, not the other way around. the "only" is misplaced
Yam Peleg@Yampeleg
You realize it's only next-token prediction? That that's ACTUALLY all it does, for real? How is any of this even real.
English
people don't find those mistakes in papers because they don't try to. i've found many such mistakes myself, they are very common. usually people just trust the proofs are right if the vibes of the result are good and the proof's surface description sounds like it would work
prinz@deredleritt3r
Sebastien Bubeck: "Already today, we're finding that [OpenAI's] models are able to surpass humans in the sense that they can find mistakes in papers. We have agents internally that have been able to find papers and say, 'hey, actually this is wrong, here is the correct answer.' Not only that, but people tend to think that AI is only good at answering questions. Actually, no, it's also pretty good at asking questions. Of course you need some research innovation there, which we had. And now our models are very good at asking questions. So good, in fact, that humans are looking at those questions and saying, 'hey, maybe I should write a paper based on this question.' So what I'm trying to say is that in a year, in two years, models could do basic[ally], more or less everything that human researchers do."
English
@Rafa_Schwinger i wouldn't be surprised if at some point an average computer could be superhuman in theorem proving
English
lmao, I swear I didn't see this paper before it
but yall know, Rafadomis will Rafadomis
Grigory Sapunov@che_shr_cat
1/ A 7B model just beat a 671B model at formal theorem proving. The secret is not more data, it is fixing the reward hacking loop in asymmetric self-play. Here is how Stanford researchers broke the RL scaling plateau. 🧵
English
Weird prediction but the best intuitive agentic engineers will have synesthesia
English
all timelines considered, the dice roll on who ends up as the "CEO of Big AI Lab" would've been the most favorable if we got someone who truly believes that AGI will make money obsolete, yet was also ideologically invested in making it available for everyone. that person would probably have invested in coding (like anthropic did), but would also make it accessible (like openai did). if we were lucky enough, perhaps we'd even get someone who'd figure out how to OSS it all without going bankrupt. that's an impossibly rare personality though. if chasing products accidentally gets us at least the availability, then I'd rather have that CEO than one who's more and more appearing to have kira complex
English
@claramilinski dentro de um meio material, vc pode pode considerar que a velocidade da luz é a velocidade de grupo, a velocidade da envoltoria do pacote, e essa velocidade pode ser praticamente zero mesmo sem reflexao. então a luz entra no modo, mas praticamente nao se propaga
Português
É sério isso? Com partículas carregadas okay, mas fóton?
GIF
Amazing Physics@amazing_physics
BREAKING🚨: For the first time in history, scientists in Italy have succeeded in "freezing" pure light and making it behave like a solid. (NATURE)
Português
@isahermanx e quando você viu o raio mas ainda nao ouviu o trovão? e se você não ver o raio mas ouvir o trovão?
Português
eu não consigo processar a ideia de que eu olho pro céu e vejo o passado
tipo a estrela que eu vejo hoje pode nem existir mais ela é só uma imagem a milhões de anos luz de delay por isso ela tá lá
se eu tivesse na estrela entao eu iria ver a terra na era dos dinossauros????
Português
@VictorTaelin i find plausible that local models can happen eventually. they just can't happen as much as the two companies can happen
English
@MacroJason i can't confirm, but i've seen someone online claim to have had access to the raw data on brazilian iq and concluded that the distribution is actually bimodal, with one of the modes corresponding to random guesses. i.e. lots of people just don't try to get the answers right
English
🇧🇷 When I posted about Brazil's low median IQ problem - many Brazilians engaged positively with my post and we had a nice exchange of ideas. Very surprised.
If I did this with nearly any other low median IQ country - I would just have tons of people crying in my replies.
Brazil is special in that it is a huge country - with illiterates and favelados but also an outsized number of bright people (more so than any other LatAm country in absolute terms). Under a different political system - they would be running things.
Jason - macro / offshore / investing@MacroJason
I like Brazil but when your population has a median numerical IQ of 83.5 - democracy and liberalism becomes a bug, not a feature. $EWZ
English
let's not forget that physics is not just theory, and in a bachelors at uni you also have courses where you do experiments. i took 5 of those, plus time working in a professor's lab. all this makes a difference.
Deivon Drago@DeivonDrago
This. 👇 I can’t imagine learning undergrad physics without all the structure. Problem sets and exams are hell, but they help you understand the foundations. Grad studies becomes much easier if you do this right.
English
asked 4.7 if an algorithm I wrote satisfies a bunch of requisites. while thinking, it realized I broke one important invariant. this was very helpful. on the final answer, though? it just said my algorithm is brilliant and flawless!!
wtf?
glad I read the thinking traces...
English
capo retweetledi
Brazil could be a superpower, except it’s missing . . . what?
English
The reason symmetry is so important in physics is because symmetry is a highly effective compression operator. If a system is invariant under some symmetry, you only need to explain one axis of it. Scientific models represent the systematic exploitation of the universe's internal redundancies through symbolic logic.
English
capo retweetledi
We should view the history of physics as a long-running program synthesis task. Kepler and Newton were searching the space of possible symbolic models to find the simplest one that would best satisfy available observations.
English
i did my master's in an experimentalist's lab, and i found extremely impressive how much the professor knew about everything: physics theory, small details about theorems, hardware, low-level programming, details about energy levels of specific chemical elements, the experimental setup itself, the literature. he basically knew everything related, from the most abstract to the most concrete level, and when trying to understand something he could think across all those levels without delay, on the fly. that was just a quantum optics lab, what happens in CERN must be absolutely insane to see
English
Physicists can seem almost insane—in the best possible way. The level of engineering and depth of physical understanding required to master these systems is far beyond what most people can imagine. It’s a combination of abstraction, intuition, and precision that feels almost unreal from the outside.
In a very real sense, this is what it looks like when humans begin to truly understand the universe. Not control it, but decode its rules at a fundamental level.
If you want a glimpse into that world, I can’t recommend enough Veritasium Youtube channel enough.
Their videos do an exceptional job of translating deep physical ideas into intuitive insights, giving you a feel for how the universe actually works beneath the surface.
Link in the comments.
English