4 papers

Evaluation & Benchmarks

Judging agents honestly: benchmarks, LLM-as-judge, and outcome measurement.

Papers
4
Read time
1.2h
Avg length
18m
01
Evaluation & Benchmarks

Ask, Don't Judge: Binary Questions for Interpretable LLM Evaluation and Self-Improvement

Instead of asking an LLM to rate an output 1–5, ask it a checklist of small yes/no questions and count the yeses — you get a score that matches humans better *and* tells you exactly what was wrong, which you can then feed back to fix either the judge or the generator.

CHO, CHAWLA, CAI, LIU, ZHU, ZHANG, SAHU (CAPITAL ONE AI FOUNDATIONS) · ICML 2026 COMPOSITIONAL LEARNING WORKSHOP
Read
18 min
Source
arXiv:2606.27226
02
Evaluation & Benchmarks

Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena

Using a strong LLM (GPT-4) as an automated judge of chatbot answers agrees with human preferences over 80% of the time — the same rate humans agree with each other — so you can replace slow, expensive human eval with a prompt, if you account for a handful of predictable biases.

ZHENG, CHIANG, SHENG, ET AL. (LMSYS / UC BERKELEY, STANFORD, CMU, UCSD) · NEURIPS 2023 DATASETS & BENCHMARKS
Read
16 min
Source
arXiv:2306.05685
03
Evaluation & Benchmarks

OSWorld: Benchmarking Multimodal Agents for Open-Ended Tasks in Real Computer Environments

OSWorld is a real, scriptable virtual-machine playground plus 369-task benchmark that lets you turn any LLM/VLM into a computer-using agent and grade it on actual desktop work — and it exposes that the best 2024 models finish only ~12% of tasks while humans hit ~72%.

TIANBAO XIE, DANYANG ZHANG, ET AL. (HKU · CMU · SALESFORCE · WATERLOO)
Read
22 min
Source
arXiv:2404.07972
04
Evaluation & Benchmarks

Outcome Primitives: A Framework for Measuring AGI Value in the World

Stop asking "can the model pass the test?" and start asking "did a real person get real work done?" — VCF gives that question a coordinate system: what kind of outcome, how big, and how sure we are it actually happened.

HAMUDI NAANAA, VITALIY SOULTAN, VOLODYMYR PANCHENKO (PORTAL AI / EVERWHY AI) · MARCH 2026
Read
14 min