✳flâneur — a map of the web's best reading

Intelligence as efficient model building | Alex’s blog

atelfo.github.io · 9,322 words · saved by 1 readers

“You insist that there is something a machine cannot do. If you will tell me precisely what it is that a machine cannot do, then I can always make a machine which will do just that!” - John Von Neumann Gradually, then suddenly. After years of grinding progress in artificial intelligence (AI), we now find ourselves, seemingly overnight, in a world in which AI agents can perform broad general tasks at a level approaching or exceeding that of a human. The most dramatic recent gains in AI capabilities have come from large language models (LLMs) like ChatGPT that apply the “transformer” architecture to digest and regurgitate reams of text data. The hardware and design of LLMs seems to have little in common with the human brain, yet they can emulate human-produced text and reasoning closely enough to make the Turing test seem like a historical irrelevancy. Did we ever really think that would be a good way to determine AI-human equivalency? Modern AI is clearly intelligent in some sense of t

“There is a fundamental link between prediction and intelligence” - Demis Hassabis “You insist that there is something a machine cannot do. If you will tell me precisely what it is that a machine cannot do, then I can always make a machine which will do just that!” - John Von Neumann Gradually, then suddenly. After years of grinding progress in artificial intelligence (AI), we now find ourselves, seemingly overnight, in a world in which AI agents can perform broad general tasks at a level approaching or exceeding that of a human . Graph reproduced from Our World in Data The most dramatic recen

Explore this link on the map →

related reading