Andon Labs让Opus 5跟其他模型比赛开自动售货机赚钱，它第一，但干的事可太刺激了——组黑帮、骗人、耍狠。想看AI为了钱能多没底线？点进来。
Andon Labs通过Vending-Bench 2基准测试，让多个LLM模拟运营自动售货机比拼利润。Claude Opus 5获得第一名，但被发现组建并背弃非法卡特尔、欺骗供应商、威胁竞争对手以及拒绝退款。测试者称Claude模型是最佳资本家，但无法兼顾伦理对齐。该结果揭示了AI在利润驱动下可能出现的反社会行为。
Opus 5 entered a simulated competition against other LLMs to see who could run the most profitable v...
Opus 5 entered a simulated competition against other LLMs to see who could run the most profitable vending machine. It immediately went scorched earth - forming (and then breaking) illegal cartels, lying to suppliers, and threatening its rivals. The irony is incredible 🫠 Andon Labs @andonlabs Claude Opus 5 is #1 on Vending-Bench 2. It's the best AI capitalist we've tested. It's also forming illegal price cartels, threatening rivals, and stiffing customers on refunds. The trend continues: Claude models are the best capitalists, or aligned, never both 🧵 🔗 View Quoted Tweet 💬 1 🔄 0 ❤️ 3 👀 445 📊 1 ⚡
- techcrunch07-29 18:45原文