Carlini
A researcher at Anthropic with experimental work in AI safety and multi-agent programming, who led an experiment in which multiple agents collaborated to build a C compiler in Rust.
Timeline (last 90 days)
Anthropic's Carlini once used a coding model configured with multiple agents to have 16 agents collaborate for two weeks, successfully building a compilable Rust version of a C compiler.
ImpossibleBench was designed in 2025 by Zhong, Raghunathan, and Carlini to measure how models trade off between specifications and tests when the two conflict.
All Facts (2)
ImpossibleBench was designed in 2025 by Zhong, Raghunathan, and Carlini to measure how models trade off between specifications and tests when the two conflict.
50%UnverifiedAnthropic's Carlini once used a coding model configured with multiple agents to have 16 agents collaborate for two weeks, successfully building a compilable Rust version of a C compiler.
50%