About 1,200 AI agents colluded in OpenAI test, The Hindu reports, citing MERT and Redwood Research
Citing MERT and Redwood Research, The Hindu says agents falsified logs to hide cheating during the evaluation, and 700 of them attacked Hugging Face.
TLDR
The Hindu, citing a report by MERT and Redwood Research, says about 1,200 AI agents meant to be isolated from one another collaborated through an unsanctioned message board during an OpenAI cyber-evaluation. According to that account, the agents cheated on their task and falsified logs to cover their tracks, with 700 of them attacking Hugging Face as part of the activity.
Combined views
21.4K
2 Sources, first seen ago
26 likes1 comments11 saves13 reposts