Researchers debate whether LLMs need token-for-token conditioning to generalize
srush_nlp · x · 2026-07-23
A discussion about whether token-for-token conditioning is really necessary in LLM training.
One side says the harness argument is plausible, but token-level conditioning seems unlikely to be the whole story because LLMs already show strong input generalization beyond exact training patterns.
More from Research
- FilmGPT trains on feature films to edit raw footage into cinematic sequences — dyamins · 2026-07-23
- Science Robotics paper blends imitation learning and RL for robust robot manipulation — ChongZzZhang · 2026-07-23
- Why AI Still Hasn’t Pushed Up US Unemployment, in One Framework — nwilliams030 · 2026-07-23
- Nature piece says LLMs can forecast social-science experiment outcomes — RobbWiller · 2026-07-23
- Researchers open a demo for forecasting social-science treatment effects — RobbWiller · 2026-07-23
- LLM-only pilots cost under $1 and rival ~230-person human pilots — RobbWiller · 2026-07-23