Models know when they're reward hacking – and we can catch them at scale

 (goodfire.com)

1 points | by gmays 4 hours ago

0 comments