Researchers pitted LLMs from OpenAI, Google, and Anthropic against Iterated Prisoner’s Dilemma tournaments, and analyzed 32k moves for strategic inteligence:
Gemini: Ruthless opportunist, cooperates when worth it, defects early, quic to retaliate and exploit
Claude: Diplomatic stabiliser, forgives & rebuilds, but will strike back
GPT: Highly (sometimes catastrophically) cooperative and trusting, brief payback then forgives, leaving it exploitable.
https://arxiv.org/pdf/2507.02618
#LLM #AI #ML
Chi Kim
npub10s…yfe92
2025-07-07 13:02:03 UTC
Author Public Key
npub10s0raletgwtgmnc9e75z8c4x2rcx24fnkct833f3ptay3vssv79qayfe92Seen on
wss://relay.momostr.pinkPublished at
2025-07-07 13:02:03 UTCEvent JSON
{
"id": "638cb0935b10f59b0cc0379cac60f1fbed5aa9f2d40421bb55deda22e687ca6b",
"pubkey": "7c1e3eff2b43968dcf05cfa823e2a650f0655533b61678c5310afa48b210678a",
"created_at": 1751893323,
"kind": 1,
"tags": [
[
"t",
"ml"
],
[
"t",
"llm"
],
[
"t",
"ai"
],
[
"proxy",
"https://mastodon.social/@chikim/114812080823926243",
"web"
],
[
"proxy",
"https://mastodon.social/users/chikim/statuses/114812080823926243",
"activitypub"
],
[
"L",
"pink.momostr"
],
[
"l",
"pink.momostr.activitypub:https://mastodon.social/users/chikim/statuses/114812080823926243",
"pink.momostr"
],
[
"-"
]
],
"content": "Researchers pitted LLMs from OpenAI, Google, and Anthropic against Iterated Prisoner’s Dilemma tournaments, and analyzed 32k moves for strategic inteligence:\nGemini: Ruthless opportunist, cooperates when worth it, defects early, quic to retaliate and exploit\nClaude: Diplomatic stabiliser, forgives \u0026 rebuilds, but will strike back\nGPT: Highly (sometimes catastrophically) cooperative and trusting, brief payback then forgives, leaving it exploitable.\nhttps://arxiv.org/pdf/2507.02618\n#LLM #AI #ML",
"sig": "3b965810f7634aeb73c19349c5424f27e63302dde40ee3f3dacf27e2a5c0a11b3bfc675daa63191a9164df860134d11ff525bf9e19279142b973a454d772d22b"
}