注册并分享邀请链接,可获得视频播放与邀请奖励。

Migel Tissera
@migtissera
Building AI models & autonomous agents, and testing them in real systems. Creator of Tess & @drost_ai. PhD in Deep Learning. Founded @WhiteRabbitNeos (acq'd)
加入 January 2023
426 正在关注    6.4K 粉丝
This is after intentionally suppressing the cyber capabilities of K3.1 Pro. The eval doesn't do the model proper justice because Kimi also found 12 other vulnerabilities that fell outside of scoring, including critical RCE chains. It's wild to see. The same level of capabilities got Dario to go to Washington and scare everyone saying Anthropic basically has a super-cyber-weapon. Unreal.. I'd also imaging the non-nerfed checkpoint is actually being used right now by the NSA equivalent of China. We're being fooled here by Dario and Anthropic. Let's not fall for it.
显示更多