Post #1790203
2026-04-15 20:44 UTC
RE: https://mathstodon.xyz/@MartinEscardo/116410439895250602
I think this bridging between communities is still useful. But it's also quite incremental work that is unlikely to lead to novel research directions.
Personally, I think we were already doing too much incremental work and too little groundbreaking work. My concern is that LLMs will tilt this imbalance even further because they are good at incremental work but horrible at groundbreaking work.
#ai #llms #science #math #physics
Replies (1)
-
@lostgen@det.social 2026-05-04 02:49
@hweimer@fediscience.org agree. LLMs are interpolation machines. They can't, by design, extrapolate. An answer that was never given is unlikely to be in the training set and hence has a very low probability. #ai #llms #science #math #physics