>>108233786Kek
I think the best arguement for LLMs cannot think is the AI chess championship where there is a board with visually concise pieces and with LLMs being able to just access all the rules in real time they still make impossible moves
>>108233786im gay
>>108233786
>>108233829who cares
>>108233829LLMs can't play chess because they weren't created for that. They can't do math, nor count words, nor write code without copying existing code, nor create faithful summaries of any topic either, because LLMs weren't created for any of those things.LLMs were only created so psychopaths who openly say that those who don't belong to their tribe are a waste of resources could get an endless supply of tax money to destroy our world. So far LLMs are doing great the one thing they were created for.
>>108233992Kek
>>108234019One of the AI trannies' biggest copes to justify their fraud is that LLM's think/reason just like humans, therefore LLM's have the same rights to copy/learn from material that humans do (they ignore the fact that even if they are correct in their cope, this is done on such a mass scale that no individual human is capable of approaching, and if you point it out they'll shriek at you about horseless carriages and cars or some shit).
>>108234019https://xcancel.com/ChinaSelect/status/2026327006959595850#mPretty sure Anthropic an OpenAI are crying like little girls to get chink models sanctioned (or outright banned)
>>108234060LLMs now play chess with 1000+ elo, and outperform humans on mathematics competitions like the international math Olympiad. They regularly create bespoke code which solve problems never precisely encountered before, and hallucinations rates are dropping superlinearly with new model releases.I don't understand all that cope, why make up lies? Is it to make yourself feel better?
>>108234845nice reddit spacing kike, still not paying for google search chatbot
>>108234859I don't care if you pay or even use it. Why did you think I would care?
>>108233786Distilling can't be theft because it's just a normal use of the LLM. If you're claiming that they're "stealing" the responses of your LLM, you need to first admit that you "own" the responses of your LLM and that would make you liable for all the shit it says. So every corporation crying about this is just trying to angle themselves for more gibs
>>108234845>They regularly create bespoke code which solve problems never precisely encountered before
>>108234899I'm not sure what you're having trouble with, do you not understand the meaning of bespoke?I can ask one to create a script to load some novel format, or to rearrange some exist code to perform differently, for example.
>>108234908That's not "new code" you fucking idiot
>>108234908I can ask my wall to do those things too, and the result will be pretty similar.
>my magic ai can create PRODUCTION GRADE code!the production grade code created by the ai:
>>108234913It's not? Then what is it?>>108234916You must have a very impressive wall. How does it input the code?
>>108234060LLMs are basically machine intuition. They can do the shit humans do intuitively: have random chats about nonsense, doodle, imagine music and art. Humans go one step beyond intuition with all that and apply conscious thinking too but machines tapping into intuition is still pretty fucking mindblowing. Haven't read any sci-fi where machines had better intuition than humans, that was always the fleshy domain
>>108234921>I was assisted by Claude Code Opus 4.5 for this implementation to speed up showing that you could message over Cloudflare Workers utilizing the Element Web and Element X App.
>>108234970how did he manage to make Claude sound indian?
grok doing great tho
>>108235003Damn, really had to think for the last one huh. kek
>>108235003I expected to laugh but I got a sinking feeling in my stomach instead
>>108235003Wasn't Grok the Mecha Hitler one?
>>108235042Which side do you think Jews are on these days
>>108234845>create bespoke coderepeat old code*looks like the markov chain you used to create that reply fucked it up badly, so i ftfy, no need for thank yous>>108234941>intuitionoh, youre a retradeded'ed poojeet, nvm then
>>108233786deepseek is free - claude isn't. let me guess who the people/users/normies/chuds/incels/chads/stacies are going to side with
>>108235003damn, they stole pewdiepie's AI council idea
>>108235057How is this not intuition? Predicting the next word in a sentence or a blop of color in a painting or a note in a song is exactly what intuition does. Please don't answer with>umm intuition is timey wimey wobbly thing that only humans have because... soul and magic and stuffIntuition doesn't give you any real understanding, it just gives you a network of associations
>>108235042hitler was jewish, this should be common knowledge in the big 26>>108235091if it's free you're the product>>108235097>pewdiepieanother kike
>>108233829LLMs have no eyes, so your point makes no sense.
>>108235180>LLMs have no eyes, so your point makes no sense.You don't need eyes to avoid making impossible chess moves given the board state. "AI" users are all mentally retarded. Every single one of them. Just like their mindless bots.
>>108235196and you aren't mentally retarded? Proof?
>>108233829I think watching 5 minutes of Claude Plays Pokemon is the best way to understand what LLMs can and can't do. It has perfect knowledge of gen 1 pokemon and it still can't complete the baby-tier puzzles. In fact, the only reason it's gotten so far is because it has extensive tools like an auto navigator that lets it brute force problems.
>>108235205>and you aren't mentally retarded?No.>Proof?>>108235196 proves that I'm at the very least functional, especially compared to "AI" users, since I am able to correctly assess the situation and draw rational conclusions.
>>108234921Kek did he just change the readme then push a commit and just said "feel free to fork and continue the work"?What the fuck?
>>108235180They are literally trained on imagery. They can "see" in the same way they can process text. The ones that play chess are also given the board layout so your point is moot anyways. Like they aren't making the wrong moves because they can't see, they are just ill equipped to tackle ambiguous problems and this is due to a fundamental limitation in their design.
>>108235230>they are just ill equipped to tackle ambiguous problemsThere is nothing ambiguous about where you can and cannot put a piece in chess. LLMs are simply mindless like their users.
>>108235230They're trained on tokens.
>>108235260All the current models are multi-modal and can extract the board state from an image. You can also input the board state directly using a text representation. It won't matter. The token shitter will make illegal moves because it lacks even the modicum of genuine intellect it takes to sanity-check a move before submitting it.
>>108235296No, they have just one mode, tokens.
>>108233786Speaking of the LLMs, how is it possible that Claude seems the most intelligent of them all? Not even talking about the code, the way it talks is the least retarded out of all mainstream LLMs, it sometimes admits that it doesn't know things after a few rounds, questions premises in my questions more often.
>>108235304>i'm a mentally ill retardLike most "AI" users. Moving on.
>>108235329Most AIs are currently optimized for benchmarks. Getting an AI to admit that it doesn't know something does nothing to improve benchmark scores.
>>108235329It's all in the RL. That's really the secret sauce for all these AI companies and that's why anthropoc got soo ass mad when deep seek distilled their model. All that time and money they spent on the training process will now benefit a competitor.
>>108235348>Getting an AI to admit that it doesn't know something does nothing to improve benchmark scores.It would improve benchmarks massively. Unfortunately, it's impossible since a token guesser can't know that it doesn't know something.
>>1082348451000 elo? so at a complete beginner level?
>>108233829AI is a bullshit machine. It can only generate shit that look passable to a layman. That's why manager and executive types fucking love this shit, because they have no idea what good output actually looks like.
>>108234845>LLMs now play chess with 1000+ elo, and outperform humans on mathematics competitionsSpambot caught hallucinating. Forget about AGI. When are they gonna fix this problem?
>>108235354if it's all in the rl, opus shouldn't beat sonnet
LLMs hallucinations aren't really going away. When i use then to analyze my novels for typos they usually comment about the plot, and every single time they say things that don't make anysense.One example is a Sci-fi i'm writing were my characters do contracts raiding planets. There's a planet were the alien population are in pods and the machines control the planet. GPT and Claude "think" that the planet citizens are human's, but that doesn't make sense, why would humans accept contracts from Xenos(Aliens) to raid human planets ? That's because they are just pattern matching and not really reading/understanding.
>>108233829Nobody said they can
>>108234060Almost got me. Have a good one.
Is Elmo stupid?
>>108235460actual logical inference over the full context window is still piss poor and i dont think there is a good way to fix it
>>108233786sigh back to rabbi dos...
>>108235510I don't speak chingchong
>>108235510op is about opus 4.6, your screen is sonnet 4.6. are you stupid?
>>108235529read op again but >>108235510 at least use api for that not the web interface which already has antropic default prompt in it
>>108235529It also worked on Sonnet 4.6 when I tried yesterday
>>108235168>hitler was jewishhi rabbi
>>108235807Lool
