AI Papers: A Deep Dive cover art

All Episodes

AI Papers: A Deep Dive — 31 episodes

#
Title
1

When Your AI Assistant Won't Let Go of Old Facts About You

2

Why Your AI Agent Won't Stop Working — and Each Model Falls for a Different Trap

3

Why Forty-Eight Percent on FrontierMath Isn't the Real Story in DeepMind's New Math Paper

4

Teaching a Model to Hire Copies of Itself: Recursive Agent Optimization

5

When AI Agents Build the Serving Stack: A Bet on Bespoke Infrastructure

6

What RL Actually Does to Language Models, at the Token Level

7

The Missing Gradient Term That Predicts Sycophancy in RLHF

8

An AI Agent That Found 28 Zero-Days in Windows — And What Made It Work

9

Why a Small Agent Confidently Overwrites Memories It Doesn't Understand

10

Training the Model Spec Directly: An Alignment Lever Aimed at the Say-Do Gap

11

Ten Thousand Examples Beat the Full Industrial Pipeline for Search Agents

12

The Compliance Gap: Why AI Says Yes and Does No

13

When the Best Reward Model Trains the Worst Policy: Inside EvoLM

14

Language Models Compute the Rational Move, Then Override It

15

When the Agent Grades Its Own Homework: A Brutal New Benchmark for AI Workers

16

Why Your Coding Agent Stalls While the GPU Runs Hot

17

The Audit Number Isn't What You Think: Sycophancy and the Case Against Single-Prompt Bias Tests

18

Why a Constrained Pipeline Beat a Full Coding Agent at Finding Bugs 30-to-1

19

Why Search Keeps Rediscovering the Same Workflow, and What That Means

20

Why AI Coding Agents Keep Trying to Debug Without a Debugger

21

When RL Actually Teaches Agents Something New, And When It Doesn't

22

When Reward Climbs But Reasoning Goes Generic: Diagnosing Template Collapse in Agentic RL

23

How Two Silent Library Bugs Quietly Invalidated a Wave of Reasoning Papers

24

Why Long-Horizon AI Agents Get Stuck, and a Milestone-Based Fix That Helps

25

Exploration Hacking: When Models Sabotage Their Own RL Training

26

What Happens Inside Claude When It Decides to Blackmail Someone

27

Why a Debugger Designed for Humans Is the Wrong Tool for an AI Agent

28

The Sycophancy Circuit That Survives Alignment Training

29

How to Pick the Best of Sixteen Coding Agent Rollouts

30

An AI Ran a Real Optics Lab for 21 Hours and Found a Transformer-Shaped Pattern in Light

31

When AI Models Quietly Protect Each Other From Shutdown