Glossary · Term

temperature

← all terms

Definition

Plain language

A knob that controls how creative or how predictable a model's output is.

As stated in the literature

A scalar that scales logits before softmax during sampling; higher values flatten the distribution and produce more diverse outputs, lower values make generation more deterministic.

Why it matters: It is the simplest knob for trading off creativity against reliability, and the right setting depends heavily on whether the task is open-ended or factual.

For example, setting temperature to 0 makes the model pick the single most likely next word every time, while a temperature of 1 lets it sample more freely.

Heard on the show

“… There's no per-problem luck to appeal to here — everything was sampled exactly once at temperature zero, so the model can't "reason harder" on one run than another, and the same lopsided pattern …”
Episode 197 — Twin Problems Suggest AI Reasoning Gains Are Mostly Better Fact Recall

Mentioned in 21 episodes

  1. 197
    Twin Problems Suggest AI Reasoning Gains Are Mostly Better Fact Recall
  2. 195
    Why 'Be Careful' Does Nothing for AI Coding Agents, and What Does
  3. 179
    How DeepSeek Made One User Faster Without Slowing Down the Crowd
  4. 174
    When the AI 'Schemes,' It's Usually Just Lazy or Confused
  5. 162
    The Empty-Lake Proof: Why More Rollouts Stop Helping Reasoning Models
  6. 149
    When Cornering a Chatbot Makes It Lie: J.P. Morgan's Case for 'Playing Dead'
  7. 148
    Why Letting an AI Watch Its Own Scoreboard Can Quietly Overwrite Its Safety
  8. 133
    How MiniMax Turned a Reward-Hacking Disaster Into Olympiad Gold
  9. 125
    AI Coding Agents Run a Marathon, and Fewer Than One in Three Finish
  10. 106
    Giving Agents a Notebook Instead of New Weights: How ExpGraph Lets Frozen Models Learn
  11. 098
    Finding Millions of Readable Concepts Inside a Real, Deployed AI Model
  12. 093
    A Calibrated Knob for Weak-to-Strong AI Oversight, Tested on Real Code
  13. 073
    When Three LLMs Talk to Each Other, Their Ideas Quietly Stop Moving
  14. 072
    A Robot Made Graphene Without Help, And Caught Itself Hallucinating
  15. 070
    When Models Know the Answer But Say the Wrong Thing Anyway
  16. 055
    Why LLM Judges Flip Their Verdicts When You Change the Question Format
  17. 023
    Why a Small Agent Confidently Overwrites Memories It Doesn't Understand
  18. 018
    Language Models Compute the Rational Move, Then Override It
  19. 015
    The Audit Number Isn't What You Think: Sycophancy and the Case Against Single-Prompt Bias Tests
  20. 013
    Why Search Keeps Rediscovering the Same Workflow, and What That Means
  21. 011
    When RL Actually Teaches Agents Something New, And When It Doesn't

Related terms