ChatGPT and Claude are ‘becoming capable of tackling real-world missions,’ say scientists

The scientists developed a tool called “AgentBench” to benchmark LLM models as agents.