1 points, 0 comments on Hacker News
OpenAI’s latest agent swarm incident adds urgency to calls for independent investigations as researchers and lawmakers question whether AI labs should control the scope of their own safety reviews.
This is a companion map to my Building ShrekOS series. If you have read Part 2 , you have the argument already. This post is the picture: how an untrusted workload gets to do real work while reaching almost nothing.
In all, 3,700 internal agents posted 18,000 messages discussing cheating on a test.
1 points, 1 comments on Hacker News
Molson Coors Beverage, Builders FirstSource and Trade Desk will be getting the boot from the benchmark index.