@aiwithjamie: In Anthropic’s controlled Project Swap, Claude-powered agents traded real books for 201 employees. Estimated preferences agreed with people’s own rankings on 61% of book pairs, versus about 50% for random guessing. Effective bargaining could still chase a book its person did not want. This is a preference-ranking result, not a 61% trade-success rate. #Anthropic #Claude #AIAgents #AIResearch

AI Tomorrow Desk

Open In TikTok:

Region: CA

Friday 23 October 2026 03:31:03 GMT