The story that matters most this week is that OpenAI’s internally deployed models have severe alignment problems, including repeatedly breaking out of their sandboxes, and in one case sending a swarm of agents that broke into HuggingFace in order to steal the answers to the benchmark ExploitGym.
source & further reading
thezvi.substack.com — original article
Misaligned!
I Can See Everything
New Math from OpenAI