Method of measuring the truthfulness of an LLM

[Paul_S]

I’m not saying this is fullproof, but it may become a defacto way of providing metrics to stakeholders about the training state of an LLM:

https://weightwatcher.ai/leaderboard.html