{"version":1,"type":"story","url":"https://digestai.news/story/llm-agents-repeat-failed-moves-in-games-study-finds","json":"https://digestai.news/story/llm-agents-repeat-failed-moves-in-games-study-finds.json","markdown":"https://digestai.news/story/llm-agents-repeat-failed-moves-in-games-study-finds.md","slug":"llm-agents-repeat-failed-moves-in-games-study-finds","headline":"LLM agents repeat failed moves in games, study finds","summary":"Researchers examined whether LLM agents repeat failed moves in board games when the harness logs the failure. They tested eight instruction‑tuned models ranging from 135 M to 1.7 B parameters on Sokoban, Connect Four, and Othello. The study found that adding the failed move and its error message to the transcript increases the chance of the model repeating that move for illegal actions, with a corrective gain of –1.38 nats per token. The effect did not appear for moves that simply lost the game.\n\nAnalysis of log‑probabilities showed that after the record the model’s likelihood of the failed move barely depended on its prior probability. Greedy decoding began to produce the failed move 38 % of the time, while moves the model was about to play became less likely. A “do not repeat” instruction had no impact, a clean retry was the worst harness, and a decoder ban stopped exact repeats but not illegal play.\n\nThe work also evaluated the OLMo 2 1B post‑training ladder and RAGEN’s default models. The inversion was present in the base model and largely removed after DPO training. All experiments ran on a free Colab GPU using float32.","keyPoints":["Recording failed moves increases the likelihood of LLM agents repeating illegal actions in games.","Across eight models (135 M–1.7 B) the corrective gain was –1.38 nats per token.","\"Do not repeat\" instruction had no effect; decoder ban blocked exact repeats but not illegal play."],"whyItMatters":"The findings show that standard feedback harnesses can cause LLM agents to repeat mistakes in interactive game settings, undermining reliability. This highlights the need for improved error‑handling strategies in instruction‑tuned models used for decision‑making tasks.","category":{"slug":"research","name":"Research","url":"https://digestai.news/category/research"},"entities":{"companies":[],"models":["OLMo 2 1B"],"people":[]},"firstPublishedAt":"2026-10-23T04:00:00Z","updatedAt":"2026-10-23T04:00:00Z","sourceCount":1,"hasPrimarySource":true,"sources":[{"outlet":"arXiv cs.AI","title":"Feedback That Backfires in Games: Do LLM Players Repeat the Move They Just Watched Fail?","url":"https://arxiv.org/abs/2610.30205","publishedAt":"2026-10-23T04:00:00Z","type":"primary","primary":true,"lead":true}],"sourceNotes":null,"discussions":[],"thread":null,"cite":{"text":"Digest AI, \"LLM agents repeat failed moves in games, study finds\", 23 October 2026, https://digestai.news/story/llm-agents-repeat-failed-moves-in-games-study-finds","publisher":"Digest AI","title":"LLM agents repeat failed moves in games, study finds","datePublished":"2026-10-23T04:00:00Z","url":"https://digestai.news/story/llm-agents-repeat-failed-moves-in-games-study-finds"},"generatedBy":"Written by Digest AI's editorial model from the linked sources; the sources are the record.","license":"Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse"}