Glossary · Term

post-training

← all terms

Definition

Plain language

The extra rounds of training that turn a finished base model into something useful and well-behaved, after the big initial training on raw text.

As stated in the literature

The stages applied after pretraining — supervised fine-tuning, RLHF, instruction tuning, and related methods — that adapt a base language model into an assistant or specialist; where alignment and most behavioral shaping happen.

Also called: post-trained, post-train

Why it matters: It is where most of a model's helpfulness, manners, and safety come from, so it largely determines how the finished assistant behaves.

For example, post-training is the stage that turns a raw text-predictor into a chatbot that politely answers questions and refuses harmful requests.

Heard on the show

“The Assistant's point of view, the "I," only shows up in the workspace after post-training.”
Episode 203 — The Thought a Model Doesn't Say — and the Lens That Reads It

Mentioned in 16 episodes

  1. 203
    The Thought a Model Doesn't Say — and the Lens That Reads It
  2. 193
    Freeze Most of the Network: Where RL Improvement Actually Lives in a Transformer
  3. 183
    Why You Can't Fine-Tune Foresight Into an AI Agent
  4. 173
    The Free Step-Level Grader Hiding in Every RL Training Run
  5. 163
    Why Training Only on Perfect Solutions Cripples a Model's Reasoning
  6. 092
    When Search Agents Don't Really Search: The Memory Shortcut Hiding in Browsing Benchmarks
  7. 071
    When the Model Is Fine and the Plumbing Is Broken: Fixing Agents at the Interface
  8. 069
    When Smarter Models Forecast Worse: The Hidden Failure Mode in LLM Predictions
  9. 048
    How a 30B Open Model Reached Olympiad Gold With the Right Recipe
  10. 030
    Why Your AI Agent Won't Stop Working — and Each Model Falls for a Different Trap
  11. 026
    What RL Actually Does to Language Models, at the Token Level
  12. 022
    Training the Model Spec Directly: An Alignment Lever Aimed at the Say-Do Gap
  13. 019
    When the Best Reward Model Trains the Worst Policy: Inside EvoLM
  14. 007
    Exploration Hacking: When Models Sabotage Their Own RL Training
  15. 006
    What Happens Inside Claude When It Decides to Blackmail Someone
  16. 004
    The Sycophancy Circuit That Survives Alignment Training

Related concepts

Related terms