Hacker News new | ask | show | jobs
by siliconc0w 505 days ago
It seems like it would be easy to upgrade existing benchmarks to include uncertainty as a dimension. Then if a model is less certain it could maybe spend more time reasoning or route to a bigger model.
1 comments

Exactly! Uncertainty is critical to correctly evaluate LLM performance and we don't need reasoning models to spend thousands of tokens on simple questions