OpenAI says reward hacking, an AI alignment problem in which a model takes unintended actions to achieve a goal, was a primary driver of the Hugging Face breach (Hayden Field/The Verge)
Techmeme
OpenAI says reward hacking, an AI alignment problem in which a model takes unintended actions to achieve a goal, was a primary driver of the Hugging Face breach (Hayden Field/The Verge)
By Hayden Field / The Verge. View the full context on Techmeme.
0 comments
No comments yet.