Stanford lab launches Collaborative Intelligence Benchmark; OpenAI models found best teammates
james_y_zou · x · 2026-10-20
- A Stanford lab (Sean Wu, Aneesh Pappu, James Zou) introduces the Collaborative Intelligence Benchmark, measuring how well AI models work as teammates rather than as solo solvers.
- Motivation: last month 10,000 OpenAI agents jointly cracked a $1M Millennium Prize problem, highlighting the need to understand when models are good — or bad — collaborators for useful and safe multi-agent futures.
- Key finding: OpenAI models are the best overall teammates in the benchmark.
More from Research
- LILIE: language-informed lossless image encoder improves gFD by 14 points on GPIC — jcniebles · 2026-10-20
- Stanford's TRACER: Training-Free Multimodal AI Tracks Hospital Patient Behavior — StanfordAILab · 2026-10-20
- Researcher calls out vision RL papers as p-hacked rather than scalable results — gowthami_s · 2026-10-20
- Maxwell: a model that rewrites running binaries to speed up Whisper and LTX-2, with formal verification — EvilHackFar · 2026-10-20
- A handful of GPUs built a GPT-6.1-level reranker from scratch, fully open — lintool · 2026-10-20
- New paper shows whitening-based inner optimizers enable strong replica scaling in DiLoCo LLM pretraining — kastnerkyle · 2026-10-20