@sameQCU I think you're wrong here. Go has a more than five times bigger branching factor than Chess. AlphaGo and successors are very much not doing "chess-engine-like" exploration.
English
FeepingCreature
5K posts
x.com/i/status/20383…
Fchollet would benefit from watching more YouTube speedrunning histories. It takes years, *years*, to study, simulate, rollout, improve strategies for simple sounding games like super Mario 64 a-press minimization. State spaces are bigger than you think.
François Chollet@fchollet
Let me explain what I mean using your chess analogy... Imagine a world where chess doesn't exist. In this world, humanity encounters an alien species, and they say "let's play a game of Glurg, it's our traditional pastime. Here are the rules, see you tomorrow" -- and it's the rules of chess. My claim is that following this interaction, a working group of the world's best minds, leveraging current externalized cognitive infrastructure (computers, the internet, etc.) would be able to analyze the rules and develop a working 3000 Elo chess engine within 24 hours, in time for the match. Give them an extra 3 weeks and they'd have a 3500 Elo engine that's 10x more compute efficient. So human intelligence is already at a level where we can go from "here are the rules" to "I can play at 3000 Elo" immediately. Not optimal yet, but not too far off.
English
@AiSimonThompson @davidad I think you can feasibly have such a mechanism in a sufficiently wide transformer forward pass. You cannot have such a thing as a *time-extended* qualia, but I suspect the brain also does not use them; rather, it instantiates qualia annotated with "we have perceived this before."
English
@FeepingCreature @davidad Consciousness is an awareness of qualia, without a mechanism to perceive the qualia how can the experience of qualia exist?
English
LLM whisperers often describe a phenomenon known as “truesight” (named after the D&D magical ability to see in total darkness). Here is some proper scientific documentation of this phenomenon.
Of course, it is not magic. It’s a combo of very strong Bayesian inference + cheating.
euan ashley@euanashley
Models performed well without, and a little better with, the images. In one case, our no-image model outperformed ALL of the current models on the chest x-ray benchmark—including the private dataset—ranking at the top of the leaderboard. Without looking at a single image. 🤯🤯
English
@Twitty_Aldo @Ubbir64 this guy fucks. there's absolutely no denying it.
English
@euanashley Have you tried yet if the vision performance increases if you exclude the samples where it is most accurate *without* vision from the training set?
That is to say, if you drop the leaky images, comparing the set of non-leaky images before and after, does it get better on them?
English
The model was shockingly good at guessing based on text clues (more than double the performance of chance alone). If you ever sat an MCQ test, you know you can use text cues to help, but, as humans, it is hard to truly comprehend the scale of knowledge embedded in these models.
English
New AI paper from us this week. When my student first showed me his initial findings, I really didn’t know what to make of them. I felt that this was an interesting but curious loophole phenomenon that would shortly be closed. I was very wrong.
arxiv.org/abs/2603.21687
English
@euanashley The answer is usually "because it wasn't trained to." Humans can have eyes opened and closed (or occluded!), so we have dedicated logic to notice when vision goes away. If you're deploying a model with vision you usually don't concern yourself with this.
English
Wait.
How come the model didn’t just say, “You forgot to show me the images”? But also, how could a visual language model do REALLY WELL on a visual task without access to any of the images?
This was wild.
🤯
English
@superkitty21 @thebestjopper I mean this is how I approach it in a tournament. When my team loses I start cheering for whoever beat them, because the further they get the less bad my team's loss looks.
English
@thebestjopper Switching teams every time your team loses sounds hilarious. Or having 32 jerseys in your closet ready to whip out at any time. Or only liking a player after they win the Hart/Art Ross and dropping them immediately when they go down in the rankings.
English
I think about this frequently
English
@AiSimonThompson @davidad I don't think consciousness is what you think it is. neither novels nor tweets have computation. I don't think enduring state is necessary for consciousness; in fact, I don't think consciousness even *has* enduring state- it's an illusion created by the nature of qualia.
English
@FeepingCreature @davidad there is no mechanism for them to be conscious; they reset on every iteration. Only the text in their context creates any change... so if you think an LLM is consious then so must a novel be, also this tweet!
English
@AiSimonThompson @davidad whatever these things are doing, it's not "seeing" as we understand it. but it's probably some form of perception. their vision is incredibly spiky, if they have a rudimentary consciousness it probably feels like blindsight. you don't see pixels, just ideas that come to mind.
English
@davidad I generated simple composites of shapes, every frontier model has problems counting them.
github.com/sgt101/llm-tes…
English
@0xextractvalue @BullMilkDealer @NathanpmYoung okay yes that is hard to tax admittedly. however from the view of the financial system it's already mostly interested in things that generate income anyways, and that's traceable.
English
@FeepingCreature @BullMilkDealer @NathanpmYoung Yeah but the taxers don't know what you are doing with income post tax. That's what you are missing. People will store their wealth post income tax in things that are impractical or impossible to tax.
I'm talking literal corporeal things like a mental ledger of favors.
English
@tenobrus Bout three hours a day, though some of that is background. Mostly gaming and discussion content.
English
i feel like i assumed everyone else in the world just kinda aged out of watching youtubers at around the same time i did but im starting to worry that may not be the case? are there still people out there who have like double digit youtube hours watched in 2026??
English
@FeepingCreature @PrinceVogel @bzogrammer this seems like a likely answer, but I like the (completely stupid) idea that there was a time in the guys' evolutionary history where it was marginally advantageous to become a soup-like homogenate for a while 😅
English
Can someone explain to the sort of selection gradient that starts an animal down the path to doing this? What does a proto-cocoon look like?
水江未来@MIRAI_MIZUE
Let's take a peek inside the pupa to see what's happening. 蛹の中で何が起こっているのか覗いてみましょう。
English
@InvaderbugGames @PrinceVogel @bzogrammer like, it's impressive in no small part because nobody would create an animal that grows like this *intentionally.*
English
@InvaderbugGames @PrinceVogel @bzogrammer that's a correct understanding! two counterarguments:
first, it doesn't necessarily need to always have done what it currently does. you can sort of mutate *into* a weird line sideways. second, why would God create an animal that grows, then ungrows, then grows again?
English
i just used the phrase ‘just a tad’ for the first time in years and thought to myself, ‘i’ve never known why tad means little, i wonder if it’s related to tadpole’ so i looked it up and it is, in fact, a shortened version of tadpole. i think that’s charming
English
@PrinceVogel @bzogrammer I want to imagine it like an embryo that can swim. you get born, you relocate somewhere more favorable, you grow up. but then evolution layered features onto the embryonic stage that had to be *unbuilt* after.
English
@bzogrammer My question is what does this look like in the proto stage? Some sort of specialized healing function?
English
@furmerly @ginnnnnn05 thesis: blowjob
antithesis: eating her out
synthesis: 69
English
@ginnnnnn05 asking the woman im having sex with if the blowjob is hegelian
English
it's been a year but someone finally found the answer to my tweet
English
@Skade590739 @NathanpmYoung I don't see why tax season would do this.
yeah country borders are a problem. you'd probably want to outlaw ownership of local assets by foreign nationals or at least mandate an in-country proxy for tax purposes. a flat tax at least makes it mathematically easy here.
English
@FeepingCreature @NathanpmYoung So like, every single tax season we just crash the valuation of any company owned by citizens of X country?
Do you not see the long term problem in that you would be catastrophically disadvantaging your own citizens' ability to invest while other nations buy up everything?
English
@ofcoralcaves @WT_Mason @NathanpmYoung "you take all the risk, the government none but share in the rewards" is literally every tax ever raised.
English
@WT_Mason @FeepingCreature @NathanpmYoung Yep. So, why bother investing? Your money has already been taxed at least once. Investment would collapse.
Why would I take all the risk & the Govt take none but share in the rewards?
Ridiculous idea.
English
@MF_Techs @pelvis_man @NathanpmYoung This is an argument about level of taxation more than kind of tax, imo.
English
@FeepingCreature @pelvis_man @NathanpmYoung Companies already pay taxes. All this does is encourage investors to move money out of the USA. It serves no other purpose.
English