Infographic: How sampling temperature flips greedy decoding vs sampling picks
cfregly · x · 2026-11-11
cfregly's illustrated series explains how temperature changes next-token selection: with the same distribution where Lori scores higher, T=0 greedy decoding deterministically picks Lori, while T=1 sampling from the original odds can unexpectedly pick Alice in a given draw.
Related event: Illustrating Sampling Temperature and Why LLM Outputs Are Non-deterministic(2 posts)→
More from Models
- Lenny: Designers name Claude most often, Anthropic winning hearts and minds — lennysan · 2026-11-11
- LLMs Aren't Fundamentally Discrete: Tokens Are Just Readouts of a Continuous Process — Sauers_ · 2026-11-11
- After Building Two Dozen Games with Claude: Its Repertoire Is Narrower Than a Human's — teortaxesTex · 2026-11-11
- Fully Local Voice-to-Voice AI on a Samsung Phone with Liquid AI's LFM2.5-Audio — alexcovo_eth · 2026-11-11
- Cartage-1: small logistics model beats frontier LLMs at 20x lower cost — shakoistsLog · 2026-11-11
- OpenAI open-sources mcp-extensions to make plugins feel native to ChatGPT — SIGKITTEN · 2026-11-11