Jade is such an excellent choice for this role! Great to have such a brilliant, wise, dedicated expert filling an important policy role.

Insightful analysis of the implications of inference scaling for AI/AGI governance. A must read.

Generate videos in just a few seconds. Try Grok Imagine, free for a limited time.

Thanks Rob for a great conversation about important topics: why technology drives history, and the rare opportunity of steering it.

RT @_lewisho: We updated our framework to include a section addressing deceptive alignment/loss of control risks. There's much more work to….

RT @ZacKenton1: We're hiring for our Google DeepMind AGI Safety & Alignment and Gemini Safety teams. Locations: London, NYC, Mountain View,….

Great work by @_lewisho, Celine Smith, Claudia van der Salm, @JoslynBarnhart, @rohinshah, @four, Jen Beroshi, @ancadianadragan, @ShaneLegg, Helen King, Tom Lue, and many others.

Many others are putting out frameworks for frontier safety, following the Seoul AI Safety Commitments. Safety scientists, policy experts, government, and industry should now identify best practice, to build standards for safe frontier AI development.

The effort involved close collaboration with our impressive security teams, in GDM and Google, to specify requisite levels for security mitigations which could be used for industry best-practice, and to map critical capability levels to security levels.

With v2 I'm especially grateful for the extensively developed framework on deceptive alignment risk, which I believe will be industry leading.

I'm proud of GoogleDeepMind/Google's v2 update to our Frontier Safety Framework. We were the first major tech company to produce an explicit risk management framework for extreme risks, and I'm glad we are continuing to push ahead on safety best practice.

Valued talking about our Frontier Safety Framework on the Responsible AI for Peace and Security Podcast. Thanks @BoulaninSIPRI

Wonderful to see the benefits of AI for science being recognized with a(nother) Nobel! Congrats to Demis, John, David, and to all working to unlock AI for science.

Seniority: we are open to hiring at virtually any seniority level. Our priority is to find excellent, driven, people who can help GDM and the world make sense of and guide frontier AI. Please apply! OR contact-frontier-safety-governance@google.com.

Geography: we have a preference for people who can join in London, though we also have hubs in NYC and the Bay Area.

How does a model's architecture, training compute, agent affordances relate to risk and appropriate guardrails? How to systematically forecast dangerous capabilities and AGI? Principled risk assessment for open weight models? Geo-strategic aspects of different kinds of compute?.

Generalist technical governance: many core issues lay in the intersection of technical ML/safety and governance, for which we would look for an agile thinker and generalist, comfortable with technical topics, enthusiasm to advance understanding. For example:.

Agent governance and safety: (semi-)autonomous agents are coming, how should they be deployed and governed? How should society manage levels of autonomy and delegated authority? How to get benefits of personalization and generality, while maintaining privacy, security, safety?.

Global and industry governance: updating global institutions and industry governance for powerful, transformative and socially beneficial AI; bringing (technical) frontier safety advice to stakeholder conversations; safety norms through FMF, AI summits, etc.

Geopolitics and AGI efforts: the path to AGI is increasingly shaped by geopolitical forces, and could see major public and private sector efforts. How will these forces evolve? Can we blueprint such an effort to be competent, safe, democracy-enhancing, globally beneficial?.

Forecasting powerful AI and AGI: work could include ML expert interviews, analyzing commissioned superforecasters, Epoch-style trend extrapolation, identification of strategic cruxes, economic analysis, decomposition and targeted forecasting of milestones to powerful AI.