We released this new agenda on LLM-safety yesterday. This is VERY comprehensive covering 18 different challenges.

My co-authors have posted tweets for each of these challenges. I am going to collect them all here!

P.S. this is also now on arxiv: arxiv.org/abs/2404.09932

1. What are the ethical and societal implications of advanced AI assistants? What might change in a world with more agentic AI?

Our new paper explores these questions:
storage.googleapis.com/deepmind-media…

It’s the result of a one year research collaboration involving 50+ researchers… a🧵

☮️🕊️

As a matter of empirical fact, we are living in a world where some say they advance AGI for benefiting all of humanity while even basic tech remains unused in distributing food to those facing famine and starvation. Who's gonna ensure that AI's advancements would benefit all of…

Detecting AI-generated deepfakes with tools like watermarks is more important than ever. 🖼️🤖
But what are watermarks, and how do they work?
Check out our latest ML & Society team blog post about AI watermarking and how it works in image, text and audio: huggingface.co/blog/watermark…

Programming is a powerful way of thinking. Your job is to enhance the productivity of thinking by a safe automation of its sub-components. It is not your job to stop a powerful way of thinking on the premise that a trillion-dollar corporation's network of computational…

Increased computational power is central to recent AI progress.

Governing compute could be an effective way to achieve AI policy goals, but could also introduce new societal risks.

We explore compute governance in a recent multi-author paper.

Today, we’re launching Aya, a new open-source, massively multilingual LLM & dataset to help support under-represented languages. Aya outperforms existing open-source models and covers 101 different languages – more than double covered by previous models.

cohere.com/research/aya

Really really interesting intervention -- strongly encourage folks working on AI existential risk to check it out

My thoughts on the fascinating new study in Science
science.org/doi/10.1126/sc… by Vong et al in the Financial Times. 1/2

ft.com/content/f67196…

🧵In the past few months, I have been working on a project involving the difficulties of doing good AI audits with black-box access. Whenever I talk about it with people, I almost always get the same question -- 'But what about security and leaks?'
arxiv.org/abs/2401.14446