3 ms·

GPT-5 set a new record on my Confabulations on Provided Texts benchmark: https://github.com/lechmazur/confabulations/ https://github.com/lechmazur/confabulation

GPT-5 set a new record on my Confabulations on Provided Texts benchmark: https://github.com/lechmazur/confabulations/ https://github.com/lechmazur/confabulations/

metzpapa 1y ago
For how much I’ve seen it pushed that this model has lower hallucination rates, it’s quite odd that every actual test I’ve seen says the opposite.
- alkyon 1y ago
  Maybe its training set included this repo?