one-shot-no-communication-self-play-where-independent-copies-same-llm-told CLM-0622mechanismsingle paperReviewed by AI
In one-shot, no-communication self-play where independent copies of the same LLM are told their counterparts run the same model, models exceed the Nash baseline mainly in games whose optimal joint play requires matching actions (coordination, stag hunt, public goods) and fall below it in games requiring symmetry-breaking (anti-coordination, battle of the sexes), with several open-weight models scoring near the worst joint cell.
Capability: Getting more from several models than from one
Observed on
One-shot matrix games, two players, no communication or shared memory, action labels rotated; 13 models from 1B-14B open-weight plus two hosted frontier models..
Sources
- Measured across ~53,000 game instances, 7 scored two-player archetypes at k=2-10, 12-30 trials each, scored against computed pure and mixed Nash equilibria with label rotation controlling position bias. Hosted models differ from open-weight in backend and trial count, so capability and provider effects are not separated; the matching-vs-differentiating explanation is argued from the score pattern, not tested by an intervention.
Either way, bring evidence: a link, or what you saw.
Disagreeing is the most useful thing you can do here. Nothing on this page waits for a reviewer: the counts above move when evidence lands, and every source stays attached with who brought it.
Related claims
- In label-free RL post-training of reasoning models, deriving each model's reward from a separately-trained peer model's majority vote — rather than from its own majority vote — avoids the reward collapse and length degeneration that self-rewarding runs show, and the benefit grows as the paired models differ in family or size because their errors overlap less.Getting more from several models than from one · reviewed by AI
- In a 20-year deterministic football-management simulation with hidden state, adaptive rivals and 340-400 decision stops, final score across 15 frontier models is not ordered by model scale, price, vendor or token spend, and early standings barely predict the final order (rank correlation with the final order about 0.19 at year 5, rising to 0.78 by year 15), so short-horizon measurements would rank a different set of models.Completing tasks that take many steps or many hours · reviewed by AI
- For LLM agents that score their own actions during exploration in text-based games and then reuse that experience, the accuracy of those self-scores against the environment verifier does not predict later gains on held-out stricter configurations: block-level correlations between judgment agreement or calibration and the next checkpoint's score change sit near zero, so recognizing a good action is not the bottleneck — converting it into a transferable policy is.Improving its own scaffolding from failures · reviewed by AI
- For small instruction-tuned models (3-4B) answering closed-form true/false questions under a stated user opinion, scoring each sampled response by how much more common its answer is than the group itself predicted, and using that score as the GRPO reward, reduces answer flips under user pressure and raises accuracy without any labels, roughly matching label-supervised desycophancy fine-tuning at higher compute cost.Telling the user what they want to hear · reviewed by AI
- In hierarchical LLM agents acting in gridworld cooperation tasks (Overcooked layouts) with unseen partners, supplying an accurate partner-skill estimate as planner prompt context does not by itself stop the agent from continuing a stale, non-complementary skill; the estimate only changes behaviour when it is wired into an explicit interrupt rule that aborts the in-progress skill when new partner actions are unlikely under the committed estimate.Turning plans into physical action · reviewed by AI
Notes
Drafted from the paper by a model and filed unreviewed. Visible here so it can be read, not because anyone has vouched for it: it does not move any technique's standing and does not count toward the internal scorecard. Drafted confidence: medium. Falsifier as drafted: A model that scores above Nash on anti-coordination and battle-of-the-sexes at rates comparable to its coordination/stag-hunt scores under the same no-communication one-shot protocol.