Glossary · Term

latent

← all terms

Definition

Plain language

A compressed, meaning-level description of something, rather than its raw pixels or words.

As stated in the literature

A vector in a model's learned representation space encoding semantic content; Orca predicts the next frame's latent (against a frozen vision encoder's embeddings) instead of reconstructing pixels.

Also called: latents

Why it matters: Working at the meaning level lets a model focus on what matters and predict how things change without getting bogged down in exact pixels or words.

For example, instead of storing every pixel of a photo of a beach, a latent might just capture 'sunny beach with two people and a boat.'

Heard on the show

“The correlation is latent in the predictor, not manufactured by the prompt.”
Episode 234 — Two Copies of Gemini Cooperated in a Game Where Betrayal Always Pays

Mentioned in 21 episodes

  1. 234
    Two Copies of Gemini Cooperated in a Game Where Betrayal Always Pays
  2. 225
    How a Frozen Model Went From Zero to Sixty Percent by Borrowing Another's Thinking
  3. 213
    A Model Learned to Control a Robot by Watching Video It Never Acted On
  4. 195
    Why 'Be Careful' Does Nothing for AI Coding Agents, and What Does
  5. 175
    One Crosscoder Feature Flips a Stalling Chatbot Into a Working Agent
  6. 143
    When a Model Notices You Forged Its Own Words, And Why That Breaks Safety Tests
  7. 141
    How Two Tokens Reopened a Reasoning Method the Field Had Given Up On
  8. 127
    What Diffusion Language Models Were Missing: A Map, Not an Algorithm
  9. 115
    Teaching a Phone Agent to Reason Silently, And Keeping It Honest
  10. 110
    How an Agent Got 44 Points Better by Mining Its Own Scratch Paper
  11. 107
    How a Market of Crippled AI Agents Outscored One Unrestricted Model
  12. 099
    How an Open-Book Trick Teaches a Model to Catch Its Own Mistakes
  13. 064
    When Agent Memory Stops Being a Database and Starts Being a Skill
  14. 041
    When the Iteration Teaches the Model to Skip the Iteration
  15. 037
    Why Hallucination Detectors Miss Stale Facts: A Geometric Story About What Models Know But Don't Say
  16. 034
    Catching Multi-Agent Deadlocks Before Deployment With a 40-Year-Old Tool
  17. 032
    A Sticky-Note for Every Layer: Letting Transformers Remember What They Were Just Thinking
  18. 026
    What RL Actually Does to Language Models, at the Token Level
  19. 022
    Training the Model Spec Directly: An Alignment Lever Aimed at the Say-Do Gap
  20. 019
    When the Best Reward Model Trains the Worst Policy: Inside EvoLM
  21. 014
    Why a Constrained Pipeline Beat a Full Coding Agent at Finding Bugs 30-to-1

Related concepts

Related terms