>/g/ insists LLMs can't do anything except predict the next token>/g/ ignores consistent gains in LLM performance and pretends like this speed of development is sub-normal instead of hyper-normal>AI-antis still quote LeCun like hes an authority on this subject despite being proven wrong several times nowWhen will /g/ antis admit they were wrong? do we really need to arrive at AGI before they will admit defeat?https://situational-awareness.ai/
>look mom the line in my graph goes UP2 more weeks etc
Look at NVIDIA revenue.
>if it can generate tokens faster then its not just generating tokensthank you for your input
>>106436848>Posts graph missing two years of data>Line is go up guys! Promise!
>2022 >Le Cunny: "i don't think we can train a machine to be intelligent purely from text"https://x.com/cammakingminds/status/1659516423540965378>ff 2025he's never admitted he was wrong btw
>>106437082btw i asked ChatGPT using LeCuns specific example of a phone on a table.
>>106437082>intelligentHe's right though
>>106437157>b-b-but standardized tests don't measure intelligence!!what do they measure then
>>106436848you're a religious freak you denounce a traditional god just to worship another made up god
>>106437166Google "intelligence"
>>106437241answer the question
>>106437222god isn't real, sweaty
>>106437290that hasn't kept anyone from fervent fanaticism
>>106436848nice, they'll be able to hallucinate slop even faster based on even more poor data
>>106437257"Intelligence" is neither defined nor measured using standardized tests. It is rather defined by an active thought proccess, and the quality of the thoughts produced, especially concerning a specific task (or the ability to comprehend/preceive more accurately, but the above is how I would expect most rational people to understand it). In humans, we cannot see the underlining thought proccess to confirm the intellect of them, thus we try to interpolate by comparing/judging the results of their thought proccess using predefined conceptions of an accurate responce.If you arent a moron, I expect you might have noticed how your question can be answered. The standardised tests do not measurw intelligence in llms simply because there is no umderlying thought proccess. Or, more generally, a test may be designed with specific preconceptions about the person taking it, the violation of which does render the test invalidSuppose we have a maze, with she solution being a set of moves (say up up right right down down right). If, in the maze, we place a trained explorer, and a kid bearing a map that clearly marks the steps to be taken to exit, the kid will most likely be the first to escape. That does not however, accurately compare the contestants path finding ability
>>106436848retarded autists who know a lot have uses
>>106437419>That does not however, accurately compare the contestants path finding abilitynta, but i struggled to find a coherent thought in all this midwitteryyou misused the word interpolate, btw
