Roger Grosse and I talk about AI safety, interpretability, regulation, registration and alignment with our colleagues Beth Coleman and Rahul Krishnan in their new podcast What Now AI?youtu.be/QDVlINfID_M?si… via YouTube

This morning, University of Toronto Engineering alum and Mozilla fellow Deborah Raji (@rajiinio) delivered the Day 2 opening keynote at SaTML Conference on the topic of audits and accountability in AI. Read a summary of Deb's remarks by Gautam Kamath below: a thread... 🧵 (1/6)

Want to know what's happening at SaTML Conference today? Luckily we have Gautam Kamath of University of Waterloo on the ground reporting for us!

First up: this morning's session on privacy chaired by @fraboeni feat. work by Ferry Julien,Junyuan Hong, and others. A thread... 🧵 #SaTML2024 (1/5)

Have you registered yet? Our annual conference, Absolutely Interdisciplinary, returns May 6-8, 2024.

This year's speakers include harper 🤯, Reza Najafi, I. Glenn Cohen, Maaz Gardezi, Gillian Hadfield, Daniel Z Buchman, Peter Loewen, Regina Rini, & more.

absolutelyinterdisciplinary.com

I don't know if we can or will build massively capable autonomous AI with long-term planning abilities. But I do know this: whether such AI gets built and deployed should be decided by us through governments, not markets. We make the case science.org/doi/10.1126/sc…

Whenever new technologies enter the world, they must earn trust. How can we create trust in AI?

A new report from a global group of experts incl. Craig Shank, Ron Bodkin, Gillian Hadfield, Dr Heidy Khlaaf (هايدي خلاف) & more explores steps towards an AI certification ecosystem.

uoft.me/ai-certificati…

The piece summarizes and expands upon a paper we published earlier this year with others, including girish sastry Miles Brundage Julian Hazell Cullen O’Keefe Gillian Hadfield

arxiv.org/abs/2402.08797

This may seem arcane but it's a big deal: European court holds that standards created by private standard-setting bodies, eg. to implement EU AI Act, must be available for free (not sold as is status quo). Rule of law, transparency and all that good stuff. infojustice.org/archives/45665

🌟 Gillian Hadfield's session on AI Governance was a game-changer! 🏛️💡 She tackled the myth of AI's inevitable growth, highlighting the need for strategic regulation and a national AI registry. A thought-provoking approach to shaping AI's future responsibly! ⚖️🤖🔗👇

A paper I am very excited to read about human-machine normativity!

My colleagues and I have two NSF-funded postdocs openings. They're at the intersections of computational social science, governance, social networks, and open source software:
recruit.ucdavis.edu/JPF06297
careers.umass.edu/amherst/en-us/…
Kindly spread the word.

Gillian Hadfield Is there an argument that more capable models increase net risk at the margin? It isn’t clear to me why focussing on regulation of compute would involve net benefit.

Glad this paper is now available: we explore the potential for using regulation of compute to govern frontier AI risks. bit.ly/CPGAI.

New NYDFS guidelines propose how insurance carriers should use AI in underwriting & pricing. InsuranceNewsNet explores the implications with commentary from SRI Policy Lead Philip Dawson who predicts a wave of similar regulation in other states.

insurancenewsnet.com/innarticle/nys…

Today, I am very proud share what we have been working on for the last 14 months. ✨

Introducing Aya -- a new state-of-art for massively multilingual models. 🔥🎉

🎉 They're live! Dive into #AIAlignment at the #AlignmentWorkshop with videos now on YouTube & our site, all with captions & transcripts. 📺 For more insights, check out our blog post. ✨Links below 🔗👇Be inspired, engage, and share your favorite insights!

We are proud to announce that SRI researcher Ishtiaque Ahmed has been named one of Microsoft Research's inaugural AI & Society Fellows! Ishtiaque will contribute to the 'Reducing the Digital Divide of Generative AI in the Global South' research challenge.

microsoft.com/en-us/research…

This is really great news

🗓️ This week: the SRI Seminar Series welcomes Dylan HadfieldMenell of MIT EECS for a talk on the dangers of aligning an agent with a single person or organization's value. Wednesday Jan. 31, 12:30pm ET.

💡Talk: “You can’t have AI safety without inclusion”
🔗uoft.me/a8x