Our new op-ed in Foreign Affairs with Jacquelyn Schneider argues against using language models for military strategy & decisions. We base our discussion on our research and how we create and train these models. Inherent limitations make them unreliable for high-stake applications.

Increasingly convinced there was a prelapsarian Tower of Babel moment where everyone used the same plug, but people angered the god of electricity and now we have this

⌛️ Only a few days left before registration fees increase for #FAccT24 ! Register by **Monday** to take advantage of early bird rates: cvent.me/xBrvwq

There's a new bill, SB-1047 'Safe and Secure Innovation for Frontier Artificial Intelligence Models Act'.

I think it could do a great deal of harm to startups, American innovation, open source, and safety. So I've written a response to the authors: 🧵
answer.ai/posts/2024-04-…

Meet Astribot S1: the Next-Gen AI Robot.

- AI-powered. Unparalleled agility, dexterity and accuracy.
- The future of AI Robot is here, and it's Naturally Yours.

➡ Astribot.com

Astribot S1: Hello World! youtu.be/AePEcHIIk9s?si… 来自 YouTube

Proud of MINT Lab PhD student JakeStone for this excellent paper with Brent Mittelstadt on ignoring legitimacy in algorithmic decisions, accepted to ACM FAccT 2024! arxiv.org/abs/2404.15680

New from me in The Guardian. Not sure AI summarisation is quite ready yet to support communicative democracy: theguardian.com/commentisfree/… (thanks to Matt Boulos for some inspiration here)

Would love this in Oz, cannot remember last time a domestic flight I booked out of CBR wasn’t delayed or cancelled.

This is incredibly interesting research, an extremely impressive undertaking. Well done Hannah Rose Kirk and team!

Today we're launching PRISM, a new resource to diversify the voices contributing to alignment. We asked 1500 people around the world for their stated preferences over LLM behaviours, then we observed their contextual preferences in 8000 convos with 21 LLMs arxiv.org/abs/2404.16019

The vast majority of people expressing concern over AI + cyber have no experience or background in cyber security. If you’re in this camp I’ve got some sobering news for you, sophisticated and low skill attackers alike are already compromising “critical infrastructure” and thats

Open AI presents The Instruction Hierarchy

Training LLMs to Prioritize Privileged Instructions

Today's LLMs are susceptible to prompt injections, jailbreaks, and other attacks that allow adversaries to overwrite a model's original instructions with their own malicious prompts.

Today at Answer.AI we've got something new for you: FSDP/QDoRA. We've tested it with AI at Meta Llama3 and the results blow away anything we've seen before.

I believe that this combination is likely to create better task-specific models than anything else at any cost. 🧵

Friends: I'm indescribably chuffed to share that my first book is finally done and out shortly!

'The Politics of Platform Regulation' 🔓🔜is my 6+yearlong empirical exploration of how governments worldwide try to shape tech's trust & safety practices...
global.oup.com/academic/produ…

We created reviewing guidelines for Conference on Language Modeling. Not intended to automate the committee work, or dictate constraints. But, to inspire a thoughtful reviewing process, for an exciting and impactful program of the highest possible quality. We have a wonderful program committee ❤️

Who will be at FAccT this year? ✨🇧🇷🙋🏻‍♀️