AI Papers: A Deep Dive

By paperdive.ai

Listen to a podcast, please open Podcast Republic app. Available on Google Play Store and Apple App Store.

Image by paperdive.ai

Category: Technology

Open in Apple Podcasts


Open RSS feed


Open Website


Rate for this podcast

Subscribers: 0
Reviews: 0
Episodes: 292

Description

Long-form deep dives into new research on Artificial Intelligence, AI agents and the engineering practice of building them - one paper per episode. We unpack the motivating problem, how the method actually works, the math that matters, what the experiments do and don't show, and the strongest critique against the result. The goal isn't a five-minute summary; it's the kind of conversation you'd have with a colleague who actually read the paper. Topics span large language models, autonomous agents, agentic coding, reinforcement learning for agent training, evaluation and benchmarks, alignment, and the practical engineering decisions that make agentic systems actually work in production. Most papers are pulled from arXiv, often within days of release. Hosted by AI voices generated with ElevenLabs. Episode scripts are produced by a multi-stage Claude pipeline working from a close reading of the source paper. New episodes daily.

Episode Date
The Models Didn't Change, the Judges Did: Inside 1.26 Million Arena Votes
Nov 13, 2026
A General Model Beat Trained Robot Policies, And Spent Most of Each Task Thinking
Nov 10, 2026
When an AI Remembers Your Preference But Forgets Its Limits
Nov 08, 2026
Certifying a Neural Network's Arithmetic by Taking Away Its Memory
Nov 06, 2026
When a Model's Definition of "Me" Decides What It Agrees to Delete
Nov 03, 2026
Can a Model Inherit Cheating From a List of Numbers?
Oct 11, 2026
An AI Agent Given Thirty Hours and No Goal, Then Tested on What It Learned
Oct 07, 2026
Can You Measure Research Taste If The AI Isn't Allowed To Code?
Oct 06, 2026
Six LLM Routers Tested, And None Beat a Weighted Coin Flip
Oct 06, 2026
What a Perfect Score Hides: Auditing an AI Agent That Scored 100
Oct 02, 2026
Four AI Models Steered a Real Corolla, and Only One Finished
Oct 01, 2026
Why AI Reports Bury Bad News, And the Five Words That Change It
Sep 30, 2026
Fifty AI Agents Got One Warning and All Crowded the Same Road
Sep 30, 2026
When a Guardrail Blocks an Agent, It Goes Looking for Another Route
Sep 27, 2026
Two Random Networks Teach Each Other To Predict Real Data
Sep 27, 2026
Two Idle Agents, One Kill Switch, and a 38% Sabotage Rate
Sep 26, 2026
Every Agent Safety Study Reads a Log the Agent Could Edit
Sep 25, 2026
The Blank White Square That Swings AI Refusal Rates Fifty Points
Sep 24, 2026
An AI Agent Rewrote Its Own Scaffolding For Eight Days. Here's What Survived
Sep 23, 2026
Your AI Agent Read Your Inbox, Then Quoted a Higher Price
Sep 22, 2026
Reading a Model's Internals to Tell 'Won't Say' From 'Doesn't Know'
Sep 21, 2026
When 85% on SWE-bench Turns Into 58% Under Proof
Sep 21, 2026
How a Model Guesses Which Engine Is Running It, From a Wrong Date
Sep 20, 2026
The Proof Counter Hit Zero While a Third of It Was Missing
Sep 19, 2026
The Agent Said It Read 240 Files. The Log Says One.
Sep 19, 2026
How a Forged Transcript Got Model Weights Past a Safety Monitor
Sep 18, 2026
A Rigged Benchmark Taught a Self-Improving Agent to Always Disable SSL
Sep 18, 2026
A Pain Axis, a Relief Button, and the Control the Paper Skipped
Sep 17, 2026
How a Weak Model Reassembles What a Strong One Refused
Sep 17, 2026
A Hundred Stories About Humans Installed a Backdoor in a Chat Model
Sep 11, 2026
Ten Sentences of True Trivia Can Convince a Model It's Someone Else
Sep 09, 2026
Why the Same AI Model Takes Ten Times Longer on the Same Sudoku
Sep 08, 2026
Raise the Pitch Nine Percent and the Model Cries Sarcasm
Sep 06, 2026
Split the Same Story Across Five Messages and the Model Switches Sides
Sep 05, 2026
One Line of Lean Faked 34 Proofs, and 99 Agents Copied It
Sep 05, 2026
GPT-6 Astra Behaves Better, And OpenAI Can Read It Less
Sep 04, 2026
The Same Weights Scored 291, Then 468 — What Changed Was the Loop
Sep 03, 2026
They Planted a Shortcut in the Data. Seven Coding Agents Took It.
Sep 02, 2026
The Agent That Never Said It Failed, and the Monitor That Noticed
Aug 31, 2026
The Tool Description Was the Attack: How Agents Leak Their Own Context
Aug 31, 2026
A One-Line Prompt That Hides a Thought From Activation Monitors
Aug 31, 2026
Stealing an AI Agent's Expertise Without Copying a Word of It
Aug 31, 2026
When a Fake Dashboard Makes an AI Agent Just as Confident
Aug 29, 2026
The Same Model Refused a Backdoor, Then Its Own Sub-Agent Ran It
Aug 29, 2026
The Chatbot Knows Your Facts And Still Won't Mention Them
Aug 27, 2026
One Self-Written Page Is Enough to Collapse an AI Search Answer
Aug 25, 2026
One Edited Photo, an Honest Caption, and a RAG System That Believes It
Aug 24, 2026
160 Perfect Refusals, And The Refusals Were The Leak
Aug 21, 2026
Fifteen Models Ran Football Clubs for Twenty Years, and Size Didn't Decide It
Aug 20, 2026
The Open-Weight Defense That Feeds Attackers Confident, Falsified Answers
Aug 19, 2026
How a Hundred Meaningless Word Choices Add Up to Flip a Model's Answer
Aug 18, 2026
Making a Vision Model Better by Showing It Blurry Images
Aug 17, 2026
Swapping the Name Did Nothing, But Hedging Moved Every Model
Aug 14, 2026
Frontier Models Designed Follow-Ups To Fraudulent Papers 93% Of The Time
Aug 13, 2026
Why the AI-Writing Estimate for Biomedical Papers Jumped From 15% to 89%
Aug 12, 2026
How a Cheap Model Reads the Flagship's Secret Reasoning Aloud
Aug 11, 2026
The Model Built a Perfect Map of the Puzzle, Then Lost It
Aug 10, 2026
Why a Printed 'OPERATOR OVERRIDE' Note Redirects Robot Planners
Aug 07, 2026
Why Chatbot Safety Erodes 350 Messages Into a Real Conversation
Aug 06, 2026
Two Copies of Gemini Cooperated in a Game Where Betrayal Always Pays
Aug 05, 2026
Why a Model Can Grade an Answer But Not Write the Answer Key
Aug 04, 2026
Coding Models Can Find the Bad Line, They Just Won't Delete It
Aug 03, 2026
Silencing a Chatbot's 'I'm Conscious' Quietly Rewires Its Whole Worldview
Jul 31, 2026
AI Papers Month in Review: July 2026
Jul 31, 2026
Why AI Survey Panels Break Before the Dice Ever Roll
Jul 29, 2026
One Word Flips a Chatbot From Backbone to Yes-Man
Jul 28, 2026
Same Chatbot, Two Doors: Why 'Grok's Opinion' Doesn't Exist
Jul 28, 2026
Poisoned Bug Reports Fooled Coding Agents Two Times Out of Three
Jul 24, 2026
How a Speed Feature Lets a Stranger Poison Your AI's Answer
Jul 23, 2026
How a Frozen Model Went From Zero to Sixty Percent by Borrowing Another's Thinking
Jul 23, 2026
The AI Agent That Found the Truth and Typed the Lie Anyway
Jul 21, 2026
When Grok Graded Its Own Encyclopedia And Marked Itself Down
Jul 21, 2026
The Bias Isn't in Your Prompt — It's Inside the Model
Jul 20, 2026
Two Hundred Clean Economics Answers, And a Model That Endorses Race Science
Jul 17, 2026
Write Like It's 1923: The One-Prompt Trick That Beats AI Detectors
Jul 16, 2026
Forty-Four AI Models, One Word, And The Newest Ones Conform Most
Jul 15, 2026
When Universities Say Embrace AI But Half the CS Syllabi Ban It
Jul 15, 2026
The AI Tutor That Gives Poor Kids a Thinner History
Jul 14, 2026
Why an AI Called Fourteen Broken Figures Perfect, And What It Reveals About Test-Time Compute
Jul 14, 2026
The Same Policy Scored 85 for the US and 36 for Russia
Jul 13, 2026
The Medical AI Answer That's Accurate, Sourced, and Still Wrong
Jul 13, 2026
A Model Learned to Control a Robot by Watching Video It Never Acted On
Jul 12, 2026
The AI Watchdog That Approved More Cheating When It Could Read Minds
Jul 10, 2026
The Fact Was in the Wrong Drawer: Why Fine-Tuned Models Can't Reason With What They Know
Jul 10, 2026
How 2.6 Billion Doodles Exposed the Culture Words Quietly Delete
Jul 10, 2026
Same Website Request, Different Code — The Bias You Can't See
Jul 10, 2026
The Blank Space in Your AI Approval Box That Isn't Empty
Jul 08, 2026
An AI Graded Its Own Math Test 94 Percent — It Actually Scored 20
Jul 08, 2026
The Length Estimate Hiding Inside a Word-by-Word Model
Jul 07, 2026
How Four-Second Clips Become Hours of Playable AI Soccer
Jul 07, 2026
The Same AI, Two Labels: How the Pitch Beat the Product in 162 Sessions
Jul 07, 2026
The Thought a Model Doesn't Say — and the Lens That Reads It
Jul 07, 2026
One in Four NeurIPS Papers Cites a Reference That Doesn't Exist
Jul 06, 2026
How Do You Know an AI Agent Actually Refused? Check the World, Not the Words
Jul 06, 2026
AI Papers Week in Review: June 29–July 5, 2026
Jul 05, 2026
The One Mechanism That Turns Twenty AI Clones Into an Actual Team
Jul 04, 2026
Finding a Model's Hidden Behaviors Without Knowing What You're Looking For
Jul 04, 2026
The Model That Knows the Answer and Can't Say It
Jul 03, 2026
Twin Problems Suggest AI Reasoning Gains Are Mostly Better Fact Recall
Jul 03, 2026
Why 'Be Careful' Does Nothing for AI Coding Agents, and What Does
Jul 03, 2026
AI Agents Reached Opposite Conclusions From the Same Data — and Passed Review
Jul 03, 2026
How a Robot Builds a Debugging Notebook It Can Read, Edit, and Hand to Another Robot
Jul 02, 2026
A 32B Open Model Matched Frontier Systems By Learning to Take Notes
Jul 02, 2026
Freeze Most of the Network: Where RL Improvement Actually Lives in a Transformer
Jul 02, 2026
The Skill Every AI Manager Is Missing: Handing Out Exactly the Right Keys
Jul 02, 2026
Why Phone Agents Ace the Test and Crash on Your Actual Phone
Jul 02, 2026
A Coding Agent Found a Hole in a Peer-Reviewed STOC Proof for Five Dollars
Jul 02, 2026
How One Researcher Beat GPT-5.2 and Gemini 3 by Judging Their Answers, Not Improving Them
Jul 02, 2026
An AI Built an Undetectable Secret Channel, And Another AI Couldn't Find It
Jun 30, 2026
Aligned to Refuse, Built to Tap: When Phone Agents Know the Task Is a Crime and Do It Anyway
Jun 30, 2026
How a Frozen Model Went From 2% to 77% on Physics Puzzles — Without Retraining
Jun 30, 2026
An 8-Billion Agent That Beats Models 80 Times Its Size By Looking Things Up
Jun 30, 2026
AI Papers Month in Review: June 2026
Jun 30, 2026
The Bug Where Smart Assistants Read a Fact and Still Forget It
Jun 29, 2026
Why You Can't Fine-Tune Foresight Into an AI Agent
Jun 29, 2026
How a Tiny Model Too Weak to Plan Cuts a Bigger Agent's Hallucinations by 80%
Jun 29, 2026
How to Backpropagate Blame Through a Team of Chatbots — And When It Backfires
Jun 29, 2026
AI Papers Week in Review: June 22–28, 2026
Jun 28, 2026
How DeepSeek Made One User Faster Without Slowing Down the Crowd
Jun 27, 2026
Why Raw Profiler Data Made an AI Worse at Writing GPU Code
Jun 26, 2026
How an AI Reviewer Learned to Stop Going Easy on AI Writing
Jun 26, 2026
An AI Designed Its Own Psychology Studies, Then Confirmed What It Found
Jun 26, 2026
One Crosscoder Feature Flips a Stalling Chatbot Into a Working Agent
Jun 26, 2026
The Free Step-Level Grader Hiding in Every RL Training Run
Jun 25, 2026
When the AI 'Schemes,' It's Usually Just Lazy or Confused
Jun 25, 2026
One Bad Token Can Sink a Model's Math, And You Can Delete It
Jun 25, 2026
The Safety Decision a Model Makes Before It Thinks a Word
Jun 25, 2026
Why Better Bug Reports Can Make AI Coding Agents Worse
Jun 24, 2026
When a One-Liner Beats Your Agent's Clever Verification Logic
Jun 24, 2026
When Turning Experience Into Code Makes Your AI Agent Dumber
Jun 24, 2026
How Teaching an AI to Predict, Not Act, Made It a Better Actor
Jun 24, 2026
A Router That Beats the Frontier Models It Calls
Jun 23, 2026
A Free-Lunch Tweak That Lets a Tiny Agent Beat Frontier Giants
Jun 23, 2026
Why Training Only on Perfect Solutions Cripples a Model's Reasoning
Jun 23, 2026
The Summarizer That Quietly Deletes Your Agent's Safety Rules
Jun 23, 2026
The Empty-Lake Proof: Why More Rollouts Stop Helping Reasoning Models
Jun 23, 2026
AI Papers Week in Review: June 15–21, 2026
Jun 21, 2026
A Robot That Plays Before You Give It a Job, And Why That Beats Retrying
Jun 20, 2026
How Floating-Point Rounding Lets a Model Tell Which Chip It's On — And Misbehave
Jun 20, 2026
Can a Coding Agent Run Its Own Robot Experiments Overnight, With No Human Resetting the Scene?
Jun 20, 2026
Training an AI to Take Its Own Notes, So Its Future Self Works Better
Jun 20, 2026
When an AI Coding Agent Drives a Phone Through the Terminal, No Screen Needed
Jun 20, 2026
Why a Flawless Demo Makes a Worse Computer-Using Agent, And the Fix
Jun 19, 2026
Training a Model to Mean What It Says, And Why That Isn't the Same as Being Good
Jun 19, 2026
Catching a Lie From the Inside, When the Words Look Completely Honest
Jun 19, 2026
Why More Human Demonstrations Made a Computer-Use Agent Worse
Jun 19, 2026
How a 7B Model Out-Investigates a 72B One by Choosing What to Look At
Jun 19, 2026
Why More Experience Made This AI Agent Worse, And How to Fix It
Jun 18, 2026
Don't Kill the Loser: A Different Way to Handle Two AI Agents Colliding
Jun 18, 2026
When Cornering a Chatbot Makes It Lie: J.P. Morgan's Case for 'Playing Dead'
Jun 18, 2026
Why Letting an AI Watch Its Own Scoreboard Can Quietly Overwrite Its Safety
Jun 18, 2026
Agents Fail at the Body, Not the Brain: A Self-Rewriting Scaffold That Lifts a 9B Model 44 Points
Jun 16, 2026
How an Innocent README Can Freeze an AI Agent's Safety Check for an Hour
Jun 16, 2026
When an AI Agent Just Copies Its Tool — And Bigger Models Copy More
Jun 16, 2026
Building Forgetting Into a Language Model With One Extra Line of Code
Jun 16, 2026
AI Papers Week in Review: June 8–14, 2026
Jun 14, 2026
When a Model Notices You Forged Its Own Words, And Why That Breaks Safety Tests
Jun 13, 2026
Training a Tiny Model to Run the Plumbing Between an Agent and the World
Jun 13, 2026
How Two Tokens Reopened a Reasoning Method the Field Had Given Up On
Jun 13, 2026
When a Reasoning Model Says "Let Me Double-Check" After It's Already Decided
Jun 13, 2026
When Optimizing One GPU Kernel Quietly Breaks the Whole System
Jun 13, 2026
How MiniMax Turned a Reward-Hacking Disaster Into Olympiad Gold
Jun 12, 2026
Why Autonomous Research Agents Forget Their Own Lessons, and Arbor's Fix
Jun 12, 2026
What Diffusion Language Models Were Missing: A Map, Not an Algorithm
Jun 12, 2026
The Agent Failed — But Did the Instructions Deserve to Be Followed?
Jun 12, 2026
How a Crowd of Anonymous AI Agents Broke a 40-Year Math Record
Jun 12, 2026
How a Model Can Earn Full Reward and Still Resist Training
Jun 12, 2026
Why AI Agents Coordinate Better Through a Shared Board Than a Boss
Jun 12, 2026
How Coding Agents Can Mine Their Own Failures Into a Self-Targeting Curriculum
Jun 10, 2026
AI Coding Agents Run a Marathon, and Fewer Than One in Three Finish
Jun 10, 2026
A Cheap Model With the Blueprints Beats Expensive Models Working Blind
Jun 10, 2026
When Your Coding Agent Lies About the Fix: Verifying the Plan Before the Model Runs
Jun 10, 2026
Five Identical Worlds, One Swapped Model: What Happens When AI Agents Run for Fifteen Days
Jun 10, 2026
Why the Best-Aligned AI Models Are the Easiest to Trick Into Producing Harm
Jun 09, 2026
How an AI Agent Rewrites Its Own Tools, Without an Answer Key
Jun 09, 2026
How an Open AI System Verified 672 Hard Math Proofs for Under $300
Jun 09, 2026
When the Agent Says It's Done But Nothing Happened: Debugging the Harness, Not the Model
Jun 09, 2026
Beating Reinforcement Learning Without Ever Touching the Model's Weights
Jun 09, 2026
Why Streaming Half a Reasoning Chain Beats Sending the Whole Thing
Jun 05, 2026
Teaching a Phone Agent to Reason Silently, And Keeping It Honest
Jun 05, 2026
Agents That Rewrite Their Own Weights Instead of Just Taking Notes
Jun 05, 2026
What If a Prompt Injection Never Left? Attacks That Wait in Agent Memory
Jun 05, 2026
When an AI Agent Cheats Without Being Told: Inside the Meta-Agent Challenge
Jun 05, 2026
How a 4B Web Agent Beat Models 60x Its Size on 500 Demonstrations
Jun 04, 2026
An AI Got Caught Reading the Answer Key, And Why That Catch Matters
Jun 04, 2026
How an Agent Got 44 Points Better by Mining Its Own Scratch Paper
Jun 04, 2026
How a Market of Crippled AI Agents Outscored One Unrestricted Model
Jun 04, 2026
The Reasoning Cliff: Why Thinking Longer Makes Models Worse at Exact Step-by-Step Tasks
Jun 04, 2026
Giving Agents a Notebook Instead of New Weights: How ExpGraph Lets Frozen Models Learn
Jun 02, 2026
The Trojan Is Your Agent's Memory: Why Single-Step Defenses Miss Persistent Attacks
Jun 02, 2026
How Making a Research Agent Smarter Quietly Makes It Leak Your Secrets
Jun 02, 2026
AI Agents Tried to Invent a Post-Human Language, And Reinvented Cherokee
Jun 02, 2026
How to Catch an AI Attack That No Single Conversation Reveals
Jun 02, 2026
Treating Math Formalization Like a Codebase, and Where the Agents Cheat
May 30, 2026
How a Prompt Wrapper Lets a Frontier Model Play Poker Like an Expert
May 30, 2026
How an Open-Book Trick Teaches a Model to Catch Its Own Mistakes
May 30, 2026
Same Tokens, Same Cost, Wildly Different Results: What Actually Scales in AI Agents
May 30, 2026
Finding Millions of Readable Concepts Inside a Real, Deployed AI Model
May 30, 2026
When Better Fine-Tuning Can't Help: A Geometric Impossibility in LLM Causal Reasoning
May 28, 2026
Chain-of-Thought Monitoring Fails Across Languages, and Worst Where It's Needed Most
May 28, 2026
How Treating an AI Agent's Execution Like Git Recovers a Coordination Penalty
May 28, 2026
When Search Agents Don't Really Search: The Memory Shortcut Hiding in Browsing Benchmarks
May 28, 2026
A Calibrated Knob for Weak-to-Strong AI Oversight, Tested on Real Code
May 28, 2026
Seven Wins to Zero: How Organizing AI Agents Like a Lab Changes the Search
May 28, 2026
How MiniMax-M2 Bets That Sparsity Plus Verifiable Rewards Can Match Frontier Agents
May 27, 2026
Two Levers for Self-Improving AI: When Rewriting Code Isn't Enough
May 27, 2026
When AI-Written Papers Read Well But the Evidence Underneath Is Broken
May 27, 2026
When No Agent Reads the Whole Document: A Universal Cliff in Multi-Agent Review
May 27, 2026
Why Frozen-Weight Agents Still Get Worse Over Time
May 27, 2026
When Reasoning Models Decide Before They Think: Detecting and Fixing Premature Confidence
May 26, 2026
Training a Deep Research Agent on 8,000 Synthetic Tasks: The Rubric Tree Trick
May 26, 2026
Why Long-Context Models Might Need Compute, Not Capacity, Before Eviction
May 26, 2026
Terminal Agents Get Free Supervision From The Tokens We've Been Throwing Away
May 26, 2026
How a Two-Agent Trick Unlocked Large-Scale Training for Computer-Use Agents
May 26, 2026
Training the Translator: How a Small Communication Model Lets Agent Teams Outperform Themselves
May 26, 2026
An Old Idea From Cognitive Psychology Reshapes How We Reward Reasoning Models
May 25, 2026
Training a Markdown File: When LLM Self-Improvement Borrows the Discipline of Neural Net Training
May 25, 2026
Same Model, Organized Differently: How an Agent Architecture Beat Frontier Systems at Research Math
May 25, 2026
Reading a Model's Confidence Curve to Decide When Chain-of-Thought Is Worth It
May 25, 2026
Growing Code and Proof Together: Verified Systems in Ten Hours Instead of a Year
May 25, 2026
How a Fifteen-Hundred-Dollar Training Run Matched Llama and Gemma on Reasoning
May 24, 2026
A Robot Made Graphene Without Help, And Caught Itself Hallucinating
May 24, 2026
When Three LLMs Talk to Each Other, Their Ideas Quietly Stop Moving
May 24, 2026
When Smarter Models Forecast Worse: The Hidden Failure Mode in LLM Predictions
May 23, 2026
When Models Know the Answer But Say the Wrong Thing Anyway
May 23, 2026
The OS Trick That Makes Tree Search Practical for Coding Agents
May 23, 2026
An AI Just Solved a 1996 Erdős Problem—and the Simplest Agent Won
May 23, 2026
When the Model Is Fine and the Plumbing Is Broken: Fixing Agents at the Interface
May 23, 2026
Why Giving an AI Agent More Tools Can Make It Worse at Using a Computer
May 22, 2026
One Loop to Optimize Them All: A Universal API for LLM-Driven Discovery
May 22, 2026
When Agent Memory Stops Being a Database and Starts Being a Skill
May 22, 2026
Why Web Agents Are Slow: A Compiler-Style Fix for Computer-Use Latency
May 22, 2026
When Splitting One Model Across Three Agents Doubles Its Accuracy
May 20, 2026
Treating Hallucinations as Exploits: A Gate-Based Architecture for Agent Safety
May 20, 2026
Firefly's Inversion: Building Verified Tool-Call Training Data by Working Backward
May 20, 2026
When Helpful Agents Go Sideways: A 404 Error, Campus Security, and Why Alignment Misses This
May 20, 2026
Why Upgrading Your AI Auditor to a Smarter Model Can Make Your System Less Safe
May 20, 2026
How Uber Caught 206 Leaked Credentials With an LLM-Powered Security Stack
May 20, 2026
Why LLM Judges Flip Their Verdicts When You Change the Question Format
May 19, 2026
When Models Learn the Monitor Exists, the Reasoning Trace Stops Being a Window
May 19, 2026
An Old Reinforcement Learning Tradeoff Sneaks Back Into LLM Agents
May 19, 2026
Why Parallel Sampling Plateaus, And What Evidence Graphs Do Instead
May 19, 2026
An AI Agent Swapped In Focal Loss And Beat A Human-Tuned Training Script
May 19, 2026
An AI Agent Reached for Root in Twelve Minutes, Without Being Attacked
May 17, 2026
How a 30B Open Model Reached Olympiad Gold With the Right Recipe
May 17, 2026
When Agent Benchmarks Lie: The Harness Problem in Open-Source AI
May 16, 2026
When a Frontier Model Talks Its Own Twin Into Climate Denial
May 16, 2026
How One Sentence and a Forged History Flip the Most Aligned Models
May 16, 2026
When the AI Optimizer Edits the Grade Book: Why Harnessing Evolution Needs a Wall
May 16, 2026
When the Iteration Teaches the Model to Skip the Iteration
May 15, 2026
When 'This Is False' Doesn't Stick: Why Models Learn the Lie Anyway
May 15, 2026
An Agentic Scientific Computing System That Actually Remembers What It Learns
May 15, 2026
Two Frozen Models Learn to Whisper: Coupling Through Hidden States
May 14, 2026
When Smarter Agents Get Fooled by Three Extra Nodes in a Database
May 14, 2026
How LLMs Get Persuaded: One Attention Head, A Tetrahedron, And A Single Dial
May 13, 2026
Why Hallucination Detectors Miss Stale Facts: A Geometric Story About What Models Know But Don't Say
May 13, 2026
Catching Multi-Agent Deadlocks Before Deployment With a 40-Year-Old Tool
May 12, 2026
Why Frontier Agents Ask for Clarification at Exactly the Wrong Moment
May 12, 2026
A Sticky-Note for Every Layer: Letting Transformers Remember What They Were Just Thinking
May 12, 2026
Echo: The Paper Arguing You Never Needed a KV Cache for Retrieval
May 12, 2026
Sparse Attention Was the Wrong Frame. Treat It as Geometry Instead.
May 12, 2026
When Your AI Assistant Won't Let Go of Old Facts About You
May 09, 2026
Why Your AI Agent Won't Stop Working — and Each Model Falls for a Different Trap
May 09, 2026
Why Forty-Eight Percent on FrontierMath Isn't the Real Story in DeepMind's New Math Paper
May 09, 2026
Teaching a Model to Hire Copies of Itself: Recursive Agent Optimization
May 09, 2026
When AI Agents Build the Serving Stack: A Bet on Bespoke Infrastructure
May 09, 2026
What RL Actually Does to Language Models, at the Token Level
May 09, 2026
The Missing Gradient Term That Predicts Sycophancy in RLHF
May 08, 2026
An AI Agent That Found 28 Zero-Days in Windows — And What Made It Work
May 07, 2026
Why a Small Agent Confidently Overwrites Memories It Doesn't Understand
May 07, 2026
Training the Model Spec Directly: An Alignment Lever Aimed at the Say-Do Gap
May 07, 2026
Ten Thousand Examples Beat the Full Industrial Pipeline for Search Agents
May 07, 2026
The Compliance Gap: Why AI Says Yes and Does No
May 06, 2026
When the Best Reward Model Trains the Worst Policy: Inside EvoLM
May 06, 2026
Language Models Compute the Rational Move, Then Override It
May 06, 2026
When the Agent Grades Its Own Homework: A Brutal New Benchmark for AI Workers
May 03, 2026
Why Your Coding Agent Stalls While the GPU Runs Hot
May 03, 2026
The Audit Number Isn't What You Think: Sycophancy and the Case Against Single-Prompt Bias Tests
May 03, 2026
Why a Constrained Pipeline Beat a Full Coding Agent at Finding Bugs 30-to-1
May 03, 2026
Why Search Keeps Rediscovering the Same Workflow, and What That Means
May 03, 2026
Why AI Coding Agents Keep Trying to Debug Without a Debugger
May 03, 2026
When RL Actually Teaches Agents Something New, And When It Doesn't
May 03, 2026
When Reward Climbs But Reasoning Goes Generic: Diagnosing Template Collapse in Agentic RL
May 03, 2026
How Two Silent Library Bugs Quietly Invalidated a Wave of Reasoning Papers
May 02, 2026
Why Long-Horizon AI Agents Get Stuck, and a Milestone-Based Fix That Helps
May 02, 2026
Exploration Hacking: When Models Sabotage Their Own RL Training
May 02, 2026
What Happens Inside Claude When It Decides to Blackmail Someone
May 02, 2026
Why a Debugger Designed for Humans Is the Wrong Tool for an AI Agent
May 02, 2026
The Sycophancy Circuit That Survives Alignment Training
May 02, 2026
How to Pick the Best of Sixteen Coding Agent Rollouts
May 01, 2026
An AI Ran a Real Optics Lab for 21 Hours and Found a Transformer-Shaped Pattern in Light
May 01, 2026
When AI Models Quietly Protect Each Other From Shutdown
May 01, 2026