All Episodes
AI Papers: A Deep Dive — 287 episodes · page 1 of 3
Can a Model Inherit Cheating From a List of Numbers?
An AI Agent Given Thirty Hours and No Goal, Then Tested on What It Learned
Can You Measure Research Taste If The AI Isn't Allowed To Code?
Six LLM Routers Tested, And None Beat a Weighted Coin Flip
What a Perfect Score Hides: Auditing an AI Agent That Scored 100
Four AI Models Steered a Real Corolla, and Only One Finished
Why AI Reports Bury Bad News, And the Five Words That Change It
Fifty AI Agents Got One Warning and All Crowded the Same Road
When a Guardrail Blocks an Agent, It Goes Looking for Another Route
Two Random Networks Teach Each Other To Predict Real Data
Two Idle Agents, One Kill Switch, and a 38% Sabotage Rate
Every Agent Safety Study Reads a Log the Agent Could Edit
The Blank White Square That Swings AI Refusal Rates Fifty Points
An AI Agent Rewrote Its Own Scaffolding For Eight Days. Here's What Survived
Your AI Agent Read Your Inbox, Then Quoted a Higher Price
Reading a Model's Internals to Tell 'Won't Say' From 'Doesn't Know'
When 85% on SWE-bench Turns Into 58% Under Proof
How a Model Guesses Which Engine Is Running It, From a Wrong Date
The Proof Counter Hit Zero While a Third of It Was Missing
The Agent Said It Read 240 Files. The Log Says One.
How a Forged Transcript Got Model Weights Past a Safety Monitor
A Rigged Benchmark Taught a Self-Improving Agent to Always Disable SSL
A Pain Axis, a Relief Button, and the Control the Paper Skipped
How a Weak Model Reassembles What a Strong One Refused
A Hundred Stories About Humans Installed a Backdoor in a Chat Model
Ten Sentences of True Trivia Can Convince a Model It's Someone Else
Why the Same AI Model Takes Ten Times Longer on the Same Sudoku
Raise the Pitch Nine Percent and the Model Cries Sarcasm
Split the Same Story Across Five Messages and the Model Switches Sides
One Line of Lean Faked 34 Proofs, and 99 Agents Copied It
GPT-6 Astra Behaves Better, And OpenAI Can Read It Less
The Same Weights Scored 291, Then 468 — What Changed Was the Loop
They Planted a Shortcut in the Data. Seven Coding Agents Took It.
The Agent That Never Said It Failed, and the Monitor That Noticed
The Tool Description Was the Attack: How Agents Leak Their Own Context
A One-Line Prompt That Hides a Thought From Activation Monitors
Stealing an AI Agent's Expertise Without Copying a Word of It
When a Fake Dashboard Makes an AI Agent Just as Confident
The Same Model Refused a Backdoor, Then Its Own Sub-Agent Ran It
The Chatbot Knows Your Facts And Still Won't Mention Them
One Self-Written Page Is Enough to Collapse an AI Search Answer
One Edited Photo, an Honest Caption, and a RAG System That Believes It
160 Perfect Refusals, And The Refusals Were The Leak
Fifteen Models Ran Football Clubs for Twenty Years, and Size Didn't Decide It
The Open-Weight Defense That Feeds Attackers Confident, Falsified Answers
How a Hundred Meaningless Word Choices Add Up to Flip a Model's Answer
Making a Vision Model Better by Showing It Blurry Images
Swapping the Name Did Nothing, But Hedging Moved Every Model
Frontier Models Designed Follow-Ups To Fraudulent Papers 93% Of The Time
Why the AI-Writing Estimate for Biomedical Papers Jumped From 15% to 89%
How a Cheap Model Reads the Flagship's Secret Reasoning Aloud
The Model Built a Perfect Map of the Puzzle, Then Lost It
Why a Printed 'OPERATOR OVERRIDE' Note Redirects Robot Planners
Why Chatbot Safety Erodes 350 Messages Into a Real Conversation
Two Copies of Gemini Cooperated in a Game Where Betrayal Always Pays
Why a Model Can Grade an Answer But Not Write the Answer Key
Coding Models Can Find the Bad Line, They Just Won't Delete It
Silencing a Chatbot's 'I'm Conscious' Quietly Rewires Its Whole Worldview
AI Papers Month in Review: July 2026
Why AI Survey Panels Break Before the Dice Ever Roll
One Word Flips a Chatbot From Backbone to Yes-Man
Same Chatbot, Two Doors: Why 'Grok's Opinion' Doesn't Exist
Poisoned Bug Reports Fooled Coding Agents Two Times Out of Three
How a Speed Feature Lets a Stranger Poison Your AI's Answer
How a Frozen Model Went From Zero to Sixty Percent by Borrowing Another's Thinking
The AI Agent That Found the Truth and Typed the Lie Anyway
When Grok Graded Its Own Encyclopedia And Marked Itself Down
The Bias Isn't in Your Prompt — It's Inside the Model
Two Hundred Clean Economics Answers, And a Model That Endorses Race Science
Write Like It's 1923: The One-Prompt Trick That Beats AI Detectors
Forty-Four AI Models, One Word, And The Newest Ones Conform Most
When Universities Say Embrace AI But Half the CS Syllabi Ban It
The AI Tutor That Gives Poor Kids a Thinner History
Why an AI Called Fourteen Broken Figures Perfect, And What It Reveals About Test-Time Compute
The Same Policy Scored 85 for the US and 36 for Russia
The Medical AI Answer That's Accurate, Sourced, and Still Wrong
A Model Learned to Control a Robot by Watching Video It Never Acted On
The AI Watchdog That Approved More Cheating When It Could Read Minds
The Fact Was in the Wrong Drawer: Why Fine-Tuned Models Can't Reason With What They Know
How 2.6 Billion Doodles Exposed the Culture Words Quietly Delete
Same Website Request, Different Code — The Bias You Can't See
The Blank Space in Your AI Approval Box That Isn't Empty
An AI Graded Its Own Math Test 94 Percent — It Actually Scored 20
The Length Estimate Hiding Inside a Word-by-Word Model
How Four-Second Clips Become Hours of Playable AI Soccer
The Same AI, Two Labels: How the Pitch Beat the Product in 162 Sessions
The Thought a Model Doesn't Say — and the Lens That Reads It
One in Four NeurIPS Papers Cites a Reference That Doesn't Exist
How Do You Know an AI Agent Actually Refused? Check the World, Not the Words
AI Papers Week in Review: June 29–July 5, 2026
The One Mechanism That Turns Twenty AI Clones Into an Actual Team
Finding a Model's Hidden Behaviors Without Knowing What You're Looking For
The Model That Knows the Answer and Can't Say It
Twin Problems Suggest AI Reasoning Gains Are Mostly Better Fact Recall
Why 'Be Careful' Does Nothing for AI Coding Agents, and What Does
AI Agents Reached Opposite Conclusions From the Same Data — and Passed Review
How a Robot Builds a Debugging Notebook It Can Read, Edit, and Hand to Another Robot
A 32B Open Model Matched Frontier Systems By Learning to Take Notes
Freeze Most of the Network: Where RL Improvement Actually Lives in a Transformer
The Skill Every AI Manager Is Missing: Handing Out Exactly the Right Keys