Philip Dawson

We’re pouring billions into building AI but barely investing in making sure it does what we want. Talked with @KevinTFrazier and @AndrewFATHOM at @Fathom_org’s Ashby Workshops about our proposal making its way through some state legislatures to build a competitive market for AI oversight. New @scaling_laws from @lawfare: podcasts.apple.com/us/podcast/liv…

@hackylawyER great to see you back... all the very best

Friends, if there's one thing I've learned from my recent health challenges, it's that our energy and attention are precious, limited resources. So, as I gradually return to work, I plan to use this platform very sparingly to share my work and follow yours.

Armilla AI, a provider of AI risk mitigation and transfer solutions, has raised a $4.5 million seed round led by Mistral Venture Partners with participation from various insurance providers. Read the article: ow.ly/gBAp50QFbxV #AI #Insurance #Reinsurance

Armilla wants to give companies a warranty for AI tcrn.ch/49itcty

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

The Expanding Universe of Generative Models with @AndrewYNg (@DeepLearningAI), @aidangomez (@cohere), @DaphneKoller (@insitro), @ylecun (@nyuniversity), @kaifulee (@01ai_yi) and @nxthompson (@theatlantic) #wef24 twitter.com/i/broadcasts/1…

#AIAct: following record-breaking negotiations, the EU has squared the circle on the world's first comprehensive rulebook for Artificial intelligence. Here is my account of what was agreed. Once again, details might change as the text is to be fine-tuned. euractiv.com/section/artifi…

#AIAct trilogue: it's a wrap! After 36 hours of intense negotiations, EU policymakers have reached a provisional deal on the world's first AI law. Much technical work remains, and it might take weeks to consolidate the text. More details to follow on the political agreement.

magnet:?xt=urn:btih:5546272da9065eddeb6fcd7ffddeef5b75be79a7&dn=mixtral-8x7b-32kseqlen&tr=udp%3A%2F%2Fopentracker.i2p.rocks%3A6969%2Fannounce&tr=http%3A%2F%https://t.co/g0m9cEUz0T%3A80%2Fannounce RELEASE a6bbd9affe0c2725c1b7410d66833e24

#AIAct: here is my account of the key parts provisionally agreed upon in the first part of the trilogue and what critical issues remain on the table for tomorrow. euractiv.com/section/artifi…

📢 Are you a #UofT researcher whose work explores the social impacts of new technologies? Calls for 2024 Schwartz Reisman faculty and graduate fellowships are now open. Deadline: February 4, 2024. Learn more and apply: uoft.me/sri-fellowship…

AI governance is daunting, but controls do exist and one of these is assurance. What would robust AI assurance look like? SRI policy experts Phil Dawson & Isaac Gazendam explore AI assurance in new report from @StanfordCyber. The report: bit.ly/3GqMRL7 Thread below.🧵⬇️

Our Executive Director @kylecmatthews along with @PhilipDawson @PaulRSamson @CIGIonline @BJMcP and @cjtermorshuizen discuss digital priorities of the G7. #cigf2023

Armilla offers verification and warranties for enterprises using AI models   venturebeat.com/ai/armilla-off…

@mustafasuleyman @intelligence2 Great takes, @mustafasuleyman. Always reminds of @richardswhitt's work on personal AI's and governance controls from earlier days... papers.ssrn.com/sol3/papers.cf…

in the next few years, everybody is going to have their own personal AI assistant imagine having a perfect tutor, coach, research assistant in your pocket 24-7... that's what's coming great to talk at @intelligence2 youtube.com/watch?v=eJf6QP…

i keep hearing people say "we are discovering new capabilities" of LLMs, or "new capabilities emerge". that's the wrong metaphor. we are guiding, shaping, and aligning these models. by showing positive examples of desired outputs we aim to increase the likelihood that an LLM produces one type of generation vs. another. capabilities don't arise on their own. there are no emergent properities here. this implies a level of autonomy or agency that simply isn't there. language really matters here. it drives expectations and affects the way others react to these models. we need to be precise. we have a huge amount of control about what LLMs do and what they don't. these are design choices. we have to make them wisely and take responsibility for their consequences.

📰 Out now! Carnegie President Tino Cuéllar, @ghadfield, and @timoreilly's commentary discussing the need for a national registry for large AI models released yesterday. Read their thoughts on AI and how society can better understand its impacts: bit.ly/46IukWp

This would be a big step change, fundamentally at odds with the old skool culture of the tech industry. But it's the right thing to do and its time for a culture shift. We in AI should welcome third party audits. 4./

Contrary to a recent paper warning of AI "model collapse," @CohereAI CEO @aidangomezzz said he thinks the "next frontier" is synthetic data that leads to knowledge discovery. venturebeat.com/ai/cohere-ceo-…

Unfortunately, we've made the difficult decision that when Bill C-18 takes effect, we will remove Canadian news links from our Search, News and Discover products. goo.gle/3NyBswV