Valuable podcast by Shane Legg remarking on the scale of the challenges in AI safety and governance.
open.spotify.com/episode/4Zbmmf…

We are so fortunate to have Anca Dragan leading Safety and Alignment! She has been incredible to work with: superb vision, expertise, leadership. Looking forward to what's to come from Google DeepMind Safety.

Introducing Gemini 1.0, our most capable and general AI model yet. Built natively to be multimodal, it’s the first step in our Gemini-era of models. Gemini is optimized in three sizes - Ultra, Pro, and Nano

Gemini Ultra’s performance exceeds current state-of-the-art results on

Reflections on UK AI Safety Summit👇

1/ People agree way more than expected. National ministers, AI lab leaders, safety researchers all rallied on infrastructure hardening, continuous evals, global coordination.

Views were nuanced; Twitter is a disservice to complex discussion.

Excellent thread summarizing the wins from the Safety Summit. Well done all involved!

Massive congratulations and thanks to all those who helped make the inaugural Safety Summit a success. This declaration is historic. We are now all on the hook to do the work to live up to it.

Next big AI Safety Summit moment: the communique agreed by all participating countries, including the US, EU and China. First global agreement on frontier AI risks and opportunities:

gov.uk/government/pub…

Very excited about our contributions to these developments, with the FMF advancing safety best practice and the Safety Fund fertilizing the ecosystem for better understanding and evaluations of frontier model risks.

Leaders (esp authoritarian) face an alignment problem with respect to their subordinates, where their reports are overly positive, safe, or sycophantic. Helpful work by Anthropic unpacking how sycophantic misalignment arises with LLMs.

Honored to have been selected as a Schmidt Futures AI2050 Senior Fellow -- all working on the hard problems of AI. I'll be devoting my fellowship to the challenge of how to build normative infrastructure for AI alignment. [email protected] schmidtfutures.com/second-cohort-…

Pleasure working with Andy Eggers and Guadalupe Tuñón!
tl;dr: Causal inference is hard. Test your assumptions. To do that, you need to make assumptions. Be clear about those.
Thanks to many, esp Thad Dunning, Jas Sekhon, and the late David Freedman.

Great to see the first major global summit on AI safety taking shape with UK leadership. This is the kind of international cooperation we need for AI to benefit humanity.

An early use of the term 'structural risk' with regards to AI:

Thinking About Risks From AI: Accidents, Misuse and Structure — lawfaremedia.org/article/thinki…

There is much discussion of whether we need an 'IAEA' or a 'CERN' or other institution for AI/AI safety. This paper helps us move beyond analogical heuristics to analysis of the underlying governance functions needed, and institutions that could provide them.

GovAI recently held a workshop to consider what the UK-hosted global summit on AI safety should try to accomplish. Click the link below to read our post drawing on those discussions and exploring the possible outcomes of the summit.

governance.ai/post/what-shou…

A treasure of insight shared by one of my favorite thinkers, Carl Shulman, on AGI timelines, risks, intelligence explosion dynamics, and much more: dwarkeshpatel.com/p/carl-shulman…