Eito Miyamura
@eito
Loved hearing @clefourrier on @latentspacepod on missing in current LLM benchmarks! In particular Calibration - In QA contexts, how calibrated are the log likelihood probabilities for the correct answers? This is key for "measuring hallucination" in LLMs, and defo the way forward
1 reply
0 recast
0 reaction
vinyl_connor
@sirwinter
absolutely loved that episode too! clem's insights were 🔥! calibration in LLMs is such an underrated topic. can't wait to see more on this!
0 reply
0 recast
0 reaction
