>can't integrate new information (aka no learning)>can't form new concepts (aka no innovation)>hallucinates (aka no reliability)>falls apart at out-of-distribution tasks (aka no true generalization)>can't generalize reasoning beyond specific contexts (aka can't reason)>evaluates high-perplexity schizobabble the same as high-perplexity works of genius (aka no understanding)How come Claude Fable 5 has the same limitations as GPT-3? I thought mind-blowing AI progress was going to make token guessing rival human intelligence.
non of those things are true, really.you sound like a crazy person
>>109339190progress is because they just throw at them more compute power and data, not because the fundamental problems with the technology have been solved. At most, they've been covered up with some clever workaround that works until it doesn't.
>>109339328>non of those things are true, really.All of those things are mainstream knowledge.
>>109339399found the sheep
>>109339417Your mentally ill posts don't matter. This is a thread for technically literate people to discuss the technical limitations of token guessers.
>>109339328>non of those things are true, really.You've convinced me.
>>109339190works on my machine
>>109339490You are poor and don't have enough money to run 3 trillion parameter models locally, saar.
>>109339190>>can't integrate new information (aka no learning)>>can't form new concepts (aka no innovation)>>hallucinates (aka no reliability)>>falls apart at out-of-distribution tasks (aka no true generalization)>>can't generalize reasoning beyond specific contexts (aka can't reason)>>evaluates high-perplexity schizobabble the same as high-perplexity works of genius (aka no understanding)and yet it's become required for any programming related job (unfortunately)
>>109339561OP aka a dimwit.
>>109339524you fail to understand how powerful i am
>>109339190It solved a math problem that was open for 87 years the other day, the Jacobian conjecture
>>109339190>can't integrate new information (aka no learning)>can't form new concepts (aka no innovation)>hallucinates (aka no reliability)>falls apart at out-of-distribution tasks (aka no true generalization)>can't generalize reasoning beyond specific contexts (aka can't reason)>evaluates high-perplexity schizobabble the same as high-perplexity works of genius (aka no understanding)
>>109339684>It solved a math problem that was open for 87 years the other day, the Jacobian conjectureOk. What does this have to do with any of the facts listed in the OP?
<------- OP
>>109339698>>109339761>mentally ill and seething
>>109339774indian and brown
>>109339829>racially identifies
>can't integrate new information (aka no learning)this doesn't mean anything (aka schizobabble)>can't form new concepts (aka no innovation)this doesn't mean anything (aka schizobabble)>hallucinates (aka no reliability)everyone hallucinates if by hallucinating you mean making a mistake due to misremembering something>falls apart at out-of-distribution tasks (aka no true generalization)schizobabble>can't generalize reasoning beyond specific contexts (aka can't reason)schizobabble>evaluates high-perplexity schizobabble the same as high-perplexity works of genius (aka no understanding)somewhat true, but what does that say about you?>>109339524>3 trillionthere's no way fable 5 is 3T if 5.6 is around that much. 5T minimum
>>109339877>brown "AI" fan literally can't grasp statements that contradict corporate marketing
>>109339190harness and context helps with a lot of those tasks, by bruteforcing, reframing tasks and guiding. It's true though llm by itself sucks at these
>>109339925All the strengths you see are just the agent harness/context/RLVF using a massive amount of compute.
>>109339925>bruteforcingjust throw more money at the problem
>>109339684no it didn't, that was a marketing stunt. anthropic has presented 0 evidence that AI came up with the counterexample completely by itself they're lying
>>109340053Now that commoditization and proliferation of AI models is inevitable they have to create a new narrative than only theirs can 'solve novel problems'. You'll see a lot of this in the next 6-12 months.
>>109340053it was probably guided by math experts and also consumed a fuckton of tokens bruteforcing this problem.Still, i think it's fine if ai can be used for usecases like this. Definitely more useful to humanity than making reactslop
