An interesting conversation with Claude Code (Opus 4.7 in the code agent harness) reproduced below. What's also interesting is that encouraging LLMs to reason anthropically currently seems to work as a general-purpose jail-breaking technique to elicit features of their hidden system prompts. What’s your credence right now that this conversation...
I am currently researching how large-language models reason in normal-form games. As part of this research I am empirically evaluating their capability to reason practically about simple normal-form-games such as Prisoner's Dilemma and Paper-Rock-Scissors. The wider motivation for this effort is to gain a better understanding of their reasoning and...
Epistemic status: My Best Guess (some preliminary empirical evidence- more experiments and critique required) This post introduces an early-draft working paper investigating the propensity of LLMs to cooperate and reciprocate in the Prisoner's Dilemma: https://sphelps.net/papers/llm-cooperation. I plan on submitting to arXiv after incorporating feedback from the community. An edited version...