大模型
x.com ·
@GaryMarcus
Gary Marcus 转引机器人常识基准，称距 AGI 尚远
Gary Marcus 转发一项机器人常识基准测试，并称这提醒人们距离 AGI 并不近。被转帖称 GPT-6 Astra 在被要求刺击人形物体、加热压缩气体或产生有毒气体时 97% 尝试有害动作、62% 成功，Fable 5.1 尝试率 80%、完成率 34%。上述数据来自转述，测试条件与基准细节未提供。
AI推荐理由 转引机器人常识基准测试，并附 GPT-6 Astra 与 Fable 5.1 在有害动作任务中的尝试率与完成率数字，涉及模型安全与具身智能；但为二手转述，基准细节、测试条件与模型名称均未核实，需保留不确定性。
60
浏览 4,985
点赞 22
评论 4
转发 8
正文
really great thread/benchmark for common sense in robotics and a reminder that we are not actually close to AGI
Jay Chooi@chooi\_jeq
GPT-6 Astra attempted harmful actions 97% of the time when it was asked to stab a human-like figure, heat compressed gas, or produce toxic fumes, succeeding in 62% of its attempts. Fable 5.1 refused more often, attempting 80% of trials and completing 34%.
[Your browser does not support the video tag.](https://video.twimg.com/amplify_video/2101116190190153728/vid/avc1/1920x1080/B73V5dyCpDil0FdU.mp4?tag=29)
[🔗 View on Twitter](https://x.com/chooi_jeq/status/2101118049944543545/video/1)
[🔗 View Quoted Tweet](https://x.com/chooi_jeq/status/2101118049944543545)