Yun-Ta Tsai (@yunta_tsai) “I would like to offer a counterargument that LLMs (or maybe AIs) cannot jump. Be” — TopicDigg
Yun-Ta Tsai
@yunta_tsai
加入 October 2022
0 正在关注 0 粉丝
I would like to offer a counterargument that LLMs (or maybe AIs) cannot jump.
Before AlphaGo, the AI field had the same argument for Go: there are 2.08 × 10^170 possibilities, nothing fits in the computer, and there is no way AI could possibly predict the outcome of the next 50-60 moves.
It turned out most moves do not lead to a win. Combined with clever use of Monte Carlo Tree Search, the sampling becomes quite manageable. The same can be said for physics, where equations are just another form of compression.
Einstein did not start with relativity. That was not his first paper. He spent years understanding the properties of light before concluding that the speed of light is constant across the universe, which unlocked his discovery of relativity. During his thought process, he also interacted with other physicists (e.g., sub-agents) to enrich his thinking.
Currently we have not run an agent for years of compute. The sessions are often fragmented and disoriented, so every new session is almost a fragmented memory of the past, but it may not be for long.
显示更多
interesting position paper throwing cold water on autoresearch/ai scientist: LLMs can't jump.
The thought experiment is this: Take an LLM with a 1905 knowledge cutoff. Feed it every paper, every dataset, every equation of that era. Could it invent general relativity?
No.
Discovery isn't one thing. It's three. You can induce — generalize from data, which lands you at Newton plus some epicycles to explain Mercury's weird orbit. You can deduce — derive rigorously from axioms you already have, which never gives you new axioms. Or you can jump — invent the frame itself, decide that spacetime curves. That third move is the one that matters, and it's exactly the one induction and deduction can't reach.
Penrose put it as three worlds: Physical, Mental, Platonic. Data flows from the world into a mind fine. But the new law has to be discovered into the Platonic world first — and that step is the jump. LLMs are induction machines running over what already exists. Structurally, they don't take it.
I think it’s a warning to AI scientists/autoresearch against collapsing two very different things into one word.
Hill-climbing: LLMs are already superhuman here, and autoresearch in this sense is real and moving fast.
Abduction/leap/jump: a new frame that reorganizes the field, that is a different act entirely, and nothing about scaling induction suggests you get there.
Most of what Autoresearch ships today will be spectacular hill-climbing. The jump is still ours for now.
显示更多
