Illustrating Sampling Temperature and Why LLM Outputs Are Non-deterministic

cfregly's illustrated series explains how sampling temperature affects next-token choices and why LLM outputs are inherently non-deterministic: even an 88%-probability token can lose under independent random sampling.

2026-11-11 ~ 2026-11-11 · 2 related posts

• Same prompt, same odds, different output: why LLMs are non-deterministic — cfregly · 2026-11-11
• Infographic: How sampling temperature flips greedy decoding vs sampling picks — cfregly · 2026-11-11