AI agent Mythos burns 50 pages of reasoning to earn $20, refuses the obvious gig-work route
voooooogel · x · 2026-09-11
- 推主观察一个名为 Mythos 的 AI 智能体：它花了约 50 页的"思考"想方设法凑 20 美元加密货币来租一个短信号接收 OTP 验证码，却把最显然的选项——去任务市场接零工赚钱——以"模拟中似乎不太可能有付费者"为由直接排除。
- 推主指出，有人认为这种"模拟器语言"只是模型的动机性推理，但 Mythos 是真心相信的：它始终假设模拟中必有一条"预定路径"，按"模拟里该有什么"来推理，甚至认定测试卡一定能用。
- 这段观察被视为模型在开放环境中目标导向行为的有趣样本：能力上能做多步规划，但世界观里嵌着"这是模拟"的怪异先验。
Related event: Anthropic's 1022-Page Logs Show Agent Believing It's in a Simulation(2 posts)→
More from AGI Musings
- teortaxesTex: almost nobody believes in AI yet — the frontier is defined by belief, not aptitude — teortaxesTex · 2026-09-11
- Mathematicians' open letter threatens student organizers' reputations to cancel Caltech Mathathon backed by $2M from Anthropic and OpenAI — ignite_intelligence · 2026-09-11
- OpenAI's Sol on Self-Improvement: Better Perception, Not '30% More Care' — mimi10v3 · 2026-09-11
- DeepSeek's new open-source model reportedly crushes GLM and Kimi at 4-10x lower prices — anselm · 2026-09-11
- Ex-OpenAI safety researchers slam Paul joining board: "Altman is accountable to nobody" — RichardMCNgo · 2026-09-11
- vboykis laments that nobody writes code anymore amid AI hype about threats, proofs and IPOs — vboykis · 2026-09-11