gpt-6IMAGE
论文 AgentTime：让 agent 自己估时，GPT-6 Astra 时长遵循率 63%
新论文测试 agent 控制自身运行时长与判断时间流逝的能力，在时长遵循任务上 GPT-6 Astra 以 63% 对 4% 大幅领先 Fable 5.1，还出现 Astra 选择睡觉不干活的情况。
Michael Ofengenden
@MOfengenden24 天前收录原推发布于 2026.10.08
Michael Ofengenden @MOfengenden原推发布于 2026.10.08
New paper: AgentTime. We ask agents to control their own runtime, forecast how long tasks take, and understand how much time passed. On duration-following, GPT-6 Astra beats Fable 5.1, 63% to 4%. In some runs, Astra likes to sleep instead of continuing work. W/@maksym_andr
x.com/mofengenden/status/2108221696138510481采集于 2026.10.09 10:38