Large language models can explore effectively in-context in multi-armed bandit tasks without special prompting.

Is it true that large language models can explore effectively in-context in multi-armed bandit tasks without special prompting?

Partly. The retrieved items only partially address the claim: a survey suggests LLMs can contribute to bandit-based decision-making, while two studies report that LLMs explore poorly or uncertainly in open-ended tasks, and one confirms in-context learning generally. None directly tests in-context exploration in multi-armed bandit tasks without special prompting.

Verdict: CONTESTED PARTIAL · 2 sources for, 2 against · Last checked 3 Oct 2026

The private library was unavailable for some searches in this check. Missing library results are not evidence against the claim.

Multi-Armed Bandits Meet Large Language Models. (n.d.). arxiv.org. https://arxiv.org/html/2505.13355v2

Large Language Models Think Too Fast To Explore Effectively. (2025). doi.org. https://doi.org/10.48550/arxiv.2501.18009

Yanda, C. (2025). Towards Data-Efficient and Explainable Large Language Models. doi.org. https://doi.org/10.7916/whvw-j004

LLaMA-Rider: Spurring Large Language Models to Explore the Open World. (2024). doi.org. https://doi.org/10.18653/v1/2024.findings-naacl.292