DeepMind paper: LLMs can't sample randomly on their own, only convert given seeds
rohanpaul_ai · x · 2026-10-27
A new Google DeepMind paper, "The Illusion of Stochasticity in LLMs" (COLM 2026), shows LLMs operating as agents cannot make genuinely random choices. Agents often need to sample actions at set odds—e.g., checking servers in unpredictable order—but LLMs claim to pick uniformly while repeatedly landing on favorites like 7, 42, or answer C.
Rigorous empirical analysis across model families, sizes, prompting styles, and distributions demonstrates the failure is fundamental: frontier models can convert externally provided random seeds into target distributions, but cannot map their internal probability estimates to stochastic outputs when sampling directly.
Related event: DeepMind Paper: LLMs Cannot Sample Randomly Without External Seeds(2 posts)→
More from coding & agent
- Dev argues there's still alpha in working with your coding agent interactively — alex_frantic · 2026-10-27
- Researchers launch AMA on patterns for safer AI agents used by citizen developers — Loose_Rip359 · 2026-10-27
- Gradio Adds In-Chat Human-in-the-Loop So ML Intern Pauses Before Training — Gradio · 2026-10-27
- Building Halo Infinite Forge maps with local LLMs, DCjson and Blender MCP — swagonflyyyy · 2026-10-27
- Ex-Unity dev Lucas Meijer says OpenClaw looks cleaned up and worth another run — lucasmeijer · 2026-10-27
- Agent skills started contradicting each other after months — how to keep them clean? — Successful-Tax1306 · 2026-10-27