Today, we’re proud to share the next iteration of our Frontier Safety Framework. The Framework has been a useful tool that’s guided our approach to assessing risk from powerful frontier AI models. Over the last year, continued collaboration with experts, further research, and the experience of applying the Framework directly to models like Gemini 2.0 has helped to deepen our understanding. We’ve used those learnings to develop this latest iteration of the Framework which touches on key areas including security mitigations and an industry-first approach to deceptive alignment. Read more about this work: https://lnkd.in/etSd5VHa
SVP at FuriosaAI | Co-Founder at NETINT | Technology Emmy Winner
1 个月AI safety remains one of the most pressing challenges in our field. It’s great to see this framework evolving with real-world learnings from models like Gemini 2.0. Looking forward to diving into the details!