Project APE launches CRED to test whether LLMs can verify research errors
soumitrashukla9 · x · 2026-07-22
Project APE introduces CRED for testing autonomous verification of research errors
The post shares a new paper, Verifying the Verifiers: Towards Autonomous Policy Evaluation, and a new benchmark called CRED.
- The authors ask whether LLMs can automatically verify key errors in policy-evaluation research as generation of plausible papers becomes cheap.
- They evaluate 60+ LLMs on the benchmark.
- The accompanying chart shows a cost-recall frontier across model release dates, comparing systems such as Mistral NeMo 12B, DeepSeek V4 Pro/Flash, Hunyuan 3, Kimi K2.6, GLM 5.2, and Codex 5.6 Luna high.
- The broader goal of Project APE is to build a system that can learn to do policy evaluation autonomously, making verification cheaper and more reliable at scale.
Related event: Project APE Evaluates LLMs as Autonomous Research Error Verifiers(5 posts)→
More from Research
- Open-source decision models sold their own base in Red Alert 2 — then learned until no human could beat them — heathercmiller · 2026-10-27
- DeepMind paper: LLMs can't sample randomly on their own — pass random numbers from code instead — rohanpaul_ai · 2026-10-27
- World Model Assigns Weight and Friction to Objects, Then Simulates with a Physics Engine — jonstephens85 · 2026-10-27
- Token Collapse: Study Finds Graph Soft Prompting Tokens in LLMs Fail to Encode Structure — Sapienza · 2026-10-27
- Mathathon 2.0 launches with Terence Tao's SAIR, $1.5M in compute for teams — _sathvikr · 2026-10-27
- The 'sweet lesson' is real: two kinds of ML problems need opposite strategies — mike64_t · 2026-10-27