gradient

Definition

Plain language

The direction to nudge a model's weights to make it do better next time.

Also called: gradients

Why it matters: It's the central signal that drives essentially all neural-network training, and nearly every optimization technique is about computing or shaping gradients better.

For example, after a model misclassifies a cat photo as a dog, the gradient tells the optimizer which weights to nudge up or down to make 'cat' more likely next time.

Heard on the show

“There's one recent image-only line in visual-document retrieval, but its main attack leans on gradient-based optimization, and that degrades badly under black-box transfer.”

Mentioned in 59 episodes

1. 247
   One Edited Photo, an Honest Caption, and a RAG System That Believes It
2. 244
   The Open-Weight Defense That Feeds Attackers Confident, Falsified Answers
3. 236
   Why a Printed 'OPERATOR OVERRIDE' Note Redirects Robot Planners
4. 226
   How a Speed Feature Lets a Stranger Poison Your AI's Answer
5. 212
   The Fact Was in the Wrong Drawer: Why Fine-Tuned Models Can't Reason With What They Know
6. 186
   How a Frozen Model Went From 2% to 77% on Physics Puzzles — Without Retraining
7. 181
   How to Backpropagate Blame Through a Team of Chatbots — And When It Backfires
8. 177
   Why Raw Profiler Data Made an AI Worse at Writing GPU Code
9. 167
   How Teaching an AI to Predict, Not Act, Made It a Better Actor
10. 166
    A Router That Beats the Frontier Models It Calls
11. 164
    The Summarizer That Quietly Deletes Your Agent's Safety Rules
12. 163
    Why Training Only on Perfect Solutions Cripples a Model's Reasoning
13. 162
    The Empty-Lake Proof: Why More Rollouts Stop Helping Reasoning Models
14. 152
    Training a Model to Mean What It Says, And Why That Isn't the Same as Being Good
15. 149
    When Cornering a Chatbot Makes It Lie: J.P. Morgan's Case for 'Playing Dead'
16. 148
    Why Letting an AI Watch Its Own Scoreboard Can Quietly Overwrite Its Safety
17. 145
    Building Forgetting Into a Language Model With One Extra Line of Code
18. 141
    How Two Tokens Reopened a Reasoning Method the Field Had Given Up On
19. 133
    How MiniMax Turned a Reward-Hacking Disaster Into Olympiad Gold
20. 128
    How a Model Can Earn Full Reward and Still Resist Training
21. 127
    What Diffusion Language Models Were Missing: A Map, Not an Algorithm
22. 126
    How Coding Agents Can Mine Their Own Failures Into a Self-Targeting Curriculum
23. 119
    Beating Reinforcement Learning Without Ever Touching the Model's Weights
24. 118
    Why the Best-Aligned AI Models Are the Easiest to Trick Into Producing Harm
25. 117
    How an Open AI System Verified 672 Hard Math Proofs for Under $300
26. 114
    Agents That Rewrite Their Own Weights Instead of Just Taking Notes
27. 110
    How an Agent Got 44 Points Better by Mining Its Own Scratch Paper
28. 106
    Giving Agents a Notebook Instead of New Weights: How ExpGraph Lets Frozen Models Learn
29. 096
    How Treating an AI Agent's Execution Like Git Recovers a Coordination Penalty
30. 090
    How MiniMax-M2 Bets That Sparsity Plus Verifiable Rewards Can Match Frontier Agents
31. 088
    Two Levers for Self-Improving AI: When Rewriting Code Isn't Enough
32. 087
    When No Agent Reads the Whole Document: A Universal Cliff in Multi-Agent Review
33. 084
    Terminal Agents Get Free Supervision From The Tokens We've Been Throwing Away
34. 082
    Training a Deep Research Agent on 8,000 Synthetic Tasks: The Rubric Tree Trick
35. 080
    How a Two-Agent Trick Unlocked Large-Scale Training for Computer-Use Agents
36. 074
    How a Fifteen-Hundred-Dollar Training Run Matched Llama and Gemma on Reasoning
37. 073
    When Three LLMs Talk to Each Other, Their Ideas Quietly Stop Moving
38. 067
    An AI Just Solved a 1996 Erdős Problem—and the Simplest Agent Won
39. 065
    One Loop to Optimize Them All: A Universal API for LLM-Driven Discovery
40. 060
    When Splitting One Model Across Three Agents Doubles Its Accuracy
41. 054
    When Models Learn the Monitor Exists, the Reasoning Trace Stops Being a Window
42. 053
    An AI Agent Swapped In Focal Loss And Beat A Human-Tuned Training Script
43. 051
    Why Parallel Sampling Plateaus, And What Evidence Graphs Do Instead
44. 048
    How a 30B Open Model Reached Olympiad Gold With the Right Recipe
45. 047
    When Agent Benchmarks Lie: The Harness Problem in Open-Source AI
46. 042
    An Agentic Scientific Computing System That Actually Remembers What It Learns
47. 041
    When the Iteration Teaches the Model to Skip the Iteration
48. 040
    Two Frozen Models Learn to Whisper: Coupling Through Hidden States
49. 039
    When Smarter Agents Get Fooled by Three Extra Nodes in a Database
50. 038
    How LLMs Get Persuaded: One Attention Head, A Tetrahedron, And A Single Dial
51. 033
    Echo: The Paper Arguing You Never Needed a KV Cache for Retrieval
52. 032
    A Sticky-Note for Every Layer: Letting Transformers Remember What They Were Just Thinking
53. 028
    Teaching a Model to Hire Copies of Itself: Recursive Agent Optimization
54. 025
    The Missing Gradient Term That Predicts Sycophancy in RLHF
55. 020
    The Compliance Gap: Why AI Says Yes and Does No
56. 010
    When Reward Climbs But Reasoning Goes Generic: Diagnosing Template Collapse in Agentic RL
57. 009
    How Two Silent Library Bugs Quietly Invalidated a Wave of Reasoning Papers
58. 007
    Exploration Hacking: When Models Sabotage Their Own RL Training
59. 001
    When AI Models Quietly Protect Each Other From Shutdown

Related concepts

Related terms