ablation

Definition

Turning off part of a model on purpose to see what stops working.

Removing or zeroing out a model component (a head, layer, or feature) to test whether the rest of the network still produces the original behavior.

Also called: ablations, ablating, ablated

Mentioned in 38 episodes

1. 079
   An Old Idea From Cognitive Psychology Reshapes How We Reward Reasoning Models
2. 078
   Training a Markdown File: When LLM Self-Improvement Borrows the Discipline of Neural Net Training
3. 077
   Reading a Model's Confidence Curve to Decide When Chain-of-Thought Is Worth It
4. 076
   Same Model, Organized Differently: How an Agent Architecture Beat Frontier Systems at Research Math
5. 075
   Growing Code and Proof Together: Verified Systems in Ten Hours Instead of a Year
6. 074
   How a Fifteen-Hundred-Dollar Training Run Matched Llama and Gemma on Reasoning
7. 071
   When the Model Is Fine and the Plumbing Is Broken: Fixing Agents at the Interface
8. 065
   One Loop to Optimize Them All: A Universal API for LLM-Driven Discovery
9. 064
   When Agent Memory Stops Being a Database and Starts Being a Skill
10. 060
    When Splitting One Model Across Three Agents Doubles Its Accuracy
11. 058
    Why Upgrading Your AI Auditor to a Smarter Model Can Make Your System Less Safe
12. 057
    How Uber Caught 206 Leaked Credentials With an LLM-Powered Security Stack
13. 055
    Why LLM Judges Flip Their Verdicts When You Change the Question Format
14. 054
    When Models Learn the Monitor Exists, the Reasoning Trace Stops Being a Window
15. 048
    How a 30B Open Model Reached Olympiad Gold With the Right Recipe
16. 046
    When the AI Optimizer Edits the Grade Book: Why Harnessing Evolution Needs a Wall
17. 045
    When a Frontier Model Talks Its Own Twin Into Climate Denial
18. 043
    When 'This Is False' Doesn't Stick: Why Models Learn the Lie Anyway
19. 042
    An Agentic Scientific Computing System That Actually Remembers What It Learns
20. 041
    When the Iteration Teaches the Model to Skip the Iteration
21. 040
    Two Frozen Models Learn to Whisper: Coupling Through Hidden States
22. 036
    Sparse Attention Was the Wrong Frame. Treat It as Geometry Instead.
23. 034
    Catching Multi-Agent Deadlocks Before Deployment With a 40-Year-Old Tool
24. 033
    Echo: The Paper Arguing You Never Needed a KV Cache for Retrieval
25. 032
    A Sticky-Note for Every Layer: Letting Transformers Remember What They Were Just Thinking
26. 024
    An AI Agent That Found 28 Zero-Days in Windows — And What Made It Work
27. 023
    Why a Small Agent Confidently Overwrites Memories It Doesn't Understand
28. 022
    Training the Model Spec Directly: An Alignment Lever Aimed at the Say-Do Gap
29. 021
    Ten Thousand Examples Beat the Full Industrial Pipeline for Search Agents
30. 020
    The Compliance Gap: Why AI Says Yes and Does No
31. 018
    Language Models Compute the Rational Move, Then Override It
32. 016
    Why Your Coding Agent Stalls While the GPU Runs Hot
33. 014
    Why a Constrained Pipeline Beat a Full Coding Agent at Finding Bugs 30-to-1
34. 013
    Why Search Keeps Rediscovering the Same Workflow, and What That Means
35. 012
    Why AI Coding Agents Keep Trying to Debug Without a Debugger
36. 010
    When Reward Climbs But Reasoning Goes Generic: Diagnosing Template Collapse in Agentic RL
37. 008
    Why Long-Horizon AI Agents Get Stuck, and a Milestone-Based Fix That Helps
38. 004
    The Sycophancy Circuit That Survives Alignment Training