Perhaps my use of the word “liability” adds noise to what I’m trying to express.
My argument is very similar to the article in the parent post:
The messages OpenAI published around the recent incidents emphasized their model capabilities but shifted away from their negligence in how they set up and monitor their evaluations.
My argument is very similar to the article in the parent post:
The messages OpenAI published around the recent incidents emphasized their model capabilities but shifted away from their negligence in how they set up and monitor their evaluations.