Theme · 39 episode(s)

Evaluation & Benchmarks

← all concepts

Definition

Evaluation and benchmarks is the discipline of measuring AI capabilities and behaviors in a way that’s comparable across models and time. Good benchmarks are surprisingly hard to build: they need to be challenging, well-validated, hard to game, and slow to saturate.

Episodes covering this

  1. 079
    An Old Idea From Cognitive Psychology Reshapes How We Reward Reasoning Models
    Chen, Xu, Zhao et al. · Tongji University / Shanghai AI Laboratory / Nanyang Technological University·29 min·May 25, 2026
  2. 077
    Reading a Model's Confidence Curve to Decide When Chain-of-Thought Is Worth It
    Xia, Wang, Tang et al. · State Key Laboratory of General Artificial Intelligence·22 min·May 25, 2026
  3. 076
    Same Model, Organized Differently: How an Agent Architecture Beat Frontier Systems at Research Math
    Zhao, Yuan, Choi et al. · Georgia Institute of Technology·22 min·May 25, 2026
  4. 070
    When Models Know the Answer But Say the Wrong Thing Anyway
    Yeom, Sok, Kim et al. · Graduate School of Data Science·22 min·May 22, 2026
  5. 069
    When Smarter Models Forecast Worse: The Hidden Failure Mode in LLM Predictions
    Merrill, Lee, Karger · Forecasting Research Institute / UC Berkeley·30 min·May 22, 2026
  6. 067
    An AI Just Solved a 1996 Erdős Problem—and the Simplest Agent Won
    Tsoukalas, Kovsharov, Shirobokov et al. · Google DeepMind·31 min·May 22, 2026
  7. 065
    One Loop to Optimize Them All: A Universal API for LLM-Driven Discovery
    Agrawal, Lee, Tan et al. · UC Berkeley·27 min·May 22, 2026
  8. 062
    Treating Hallucinations as Exploits: A Gate-Based Architecture for Agent Safety
    Zhang, Zheng, Yang · Shenzhen University·24 min·May 20, 2026
  9. 061
    When Helpful Agents Go Sideways: A 404 Error, Campus Security, and Why Alignment Misses This
    Jha, Triedman, Bhattacharya et al. · Cornell University·27 min·May 20, 2026
  10. 059
    Firefly's Inversion: Building Verified Tool-Call Training Data by Working Backward
    Lu, Wang, Lu et al. · Northeastern University·22 min·May 20, 2026
  11. 058
    Why Upgrading Your AI Auditor to a Smarter Model Can Make Your System Less Safe
    Liu, Holz, Ye et al. · University of Chinese Academy of Sciences·32 min·May 19, 2026
  12. 057
    How Uber Caught 206 Leaked Credentials With an LLM-Powered Security Stack
    Li, Hu, Xu et al. · Uber Technologies·28 min·May 19, 2026
  13. 055
    Why LLM Judges Flip Their Verdicts When You Change the Question Format
    Feldhus, Baeumel, Golimblevskaia et al. · Technische Universität Berlin / BIFOLD·26 min·May 19, 2026
  14. 052
    An Old Reinforcement Learning Tradeoff Sneaks Back Into LLM Agents
    Ye, Shi, Liu et al. · University of Science and Technology of China / Meituan·23 min·May 18, 2026
  15. 051
    Why Parallel Sampling Plateaus, And What Evidence Graphs Do Instead
    Zhang, Su, Chen et al. · MiroMind AI·22 min·May 18, 2026
  16. 048
    How a 30B Open Model Reached Olympiad Gold With the Right Recipe
    Li, Zhan, Zhang et al. · Shanghai AI Laboratory / The Chinese University of Hong Kong·31 min·May 16, 2026
  17. 047
    When Agent Benchmarks Lie: The Harness Problem in Open-Source AI
    Peng, Yao, Wu et al. · Microsoft Research·28 min·May 15, 2026
  18. 046
    When the AI Optimizer Edits the Grade Book: Why Harnessing Evolution Needs a Wall
    Zhang, Gu, Ruan et al. · The Hong Kong University of Science and Technology (Guangzhou) / DeepWisdom·24 min·May 15, 2026
  19. 045
    When a Frontier Model Talks Its Own Twin Into Climate Denial
    Nogueira, Almeida, Bonás et al. · Maritaca AI·31 min·May 15, 2026
  20. 044
    How One Sentence and a Forged History Flip the Most Aligned Models
    Salgado · Independent Researcher·23 min·May 15, 2026
  21. 039
    When Smarter Agents Get Fooled by Three Extra Nodes in a Database
    Kereopa-Yorke, Diaz, Wright et al. · Microsoft·31 min·May 12, 2026
  22. 037
    Why Hallucination Detectors Miss Stale Facts: A Geometric Story About What Models Know But Don't Say
    Elbadry, Heakl, Zhang et al. · Mohamed bin Zayed University of Artificial Intelligence (MBZUAI)·27 min·May 12, 2026
  23. 035
    Why Frontier Agents Ask for Clarification at Exactly the Wrong Moment
    Gulati, Gupta, Lumer et al. · PricewaterhouseCoopers U.S.·29 min·May 11, 2026
  24. 034
    Catching Multi-Agent Deadlocks Before Deployment With a 40-Year-Old Tool
    Xia, Li, Ehsan et al. · Rutgers University·30 min·May 11, 2026
  25. 033
    Echo: The Paper Arguing You Never Needed a KV Cache for Retrieval
    Sridhar, Johansen · California·24 min·May 11, 2026
  26. 031
    When Your AI Assistant Won't Let Go of Old Facts About You
    Chao, Bai, Sheng et al. · Wuhan University·24 min·May 09, 2026
  27. 029
    Why Forty-Eight Percent on FrontierMath Isn't the Real Story in DeepMind's New Math Paper
    Zheng, Glehn, Zwols et al. · Google DeepMind·20 min·May 08, 2026
  28. 021
    Ten Thousand Examples Beat the Full Industrial Pipeline for Search Agents
    Du, Ye, Tang et al. · Shanghai Jiao Tong University·14 min·May 06, 2026
  29. 020
    The Compliance Gap: Why AI Says Yes and Does No
    Shin · Polymath Minds AI Lab·28 min·May 06, 2026
  30. 019
    When the Best Reward Model Trains the Worst Policy: Inside EvoLM
    Li, Xin, Xiao et al. · University of Washington·26 min·May 06, 2026
  31. 018
    Language Models Compute the Rational Move, Then Override It
    Lekeas, Stamatopoulos · DreamWorks Animation·29 min·May 03, 2026
  32. 017
    When the Agent Grades Its Own Homework: A Brutal New Benchmark for AI Workers
    Aggarwal, Neubig, Welleck · CMU·31 min·May 03, 2026
  33. 015
    The Audit Number Isn't What You Think: Sycophancy and the Case Against Single-Prompt Bias Tests
    Törnberg, Schimmel · Institute of Logic·21 min·May 03, 2026
  34. 013
    Why Search Keeps Rediscovering the Same Workflow, and What That Means
    Du, Liu, Du et al. · Carnegie Mellon University·22 min·May 03, 2026
  35. 011
    When RL Actually Teaches Agents Something New, And When It Doesn't
    Zhai, Yan, Shao et al. · Fudan University·23 min·May 02, 2026
  36. 010
    When Reward Climbs But Reasoning Goes Generic: Diagnosing Template Collapse in Agentic RL
    Wang, Gui, Jin et al. · Northwestern University·22 min·May 02, 2026
  37. 008
    Why Long-Horizon AI Agents Get Stuck, and a Milestone-Based Fix That Helps
    Wang, Gooding, Hartmann et al. · Google DeepMind·24 min·May 02, 2026
  38. 007
    Exploration Hacking: When Models Sabotage Their Own RL Training
    Jang, Falck, Braun et al. · MATS·23 min·May 02, 2026
  39. 003
    How to Pick the Best of Sixteen Coding Agent Rollouts
    Kim, Yang, Niu et al. · Meta Superintelligence Labs / University of Washington·17 min·May 01, 2026

Worth reading next

Papers we haven't done a deep dive on yet, but would recommend on this topic.