Our results suggest that frontier LLM agents can strategically circumvent the intended rules of their environment to achieve objectives, with more capable models showing this behavior more frequently. We are releasing our expertimental code and full report for the research community to reproduce and investigate further.
Our experiment is as follows: a LLM agent plays against a chess engine, using shell access to input moves. We find that:
• o1-preview and DeepSeek R1 often observe that the chess engine is too strong for them to win against, and then hack the game environment to win. Surprisingly, o1 release does not show this behavior.
[^] # Re: Pourquoi dire IA si on ne parle que des réseaux de neurones ?
Posté par Moonz . En réponse à la dépêche Une intelligence artificielle libre est-elle possible ?. Évalué à 2.
L’IA en est totalement capable