🚨Our paper on identifying principles of justice for AI is out in PNAS today!🚨 Participants who choose principles from behind a “veil of ignorance” prioritise the worst-off, even where this does not benefit them directly. What this means for AI ethics 👇🧵

our @deepmind paper is out in PNAS! we use john rawls' veil of ignorance as a method of choosing principles to align AI 👰‍♀️🤖 participants behind the veil are more likely to choose principles that prioritize the worst off, driven by fairness concerns https://t.co/fObELrlegW

We @wellcometrust are trying to develop our understanding of how #mentalhealth lived experience can inform the work of basic neuroscientists. If you have any learnings you would be happy to share please get in touch with the excellent @maisiejenk

thanks to everyone for your interest in this book. it has now been published.

Should I step down as head of Twitter? I will abide by the results of this poll.

This is very cool work in a direction that is critically important: better techniques for eliciting the values that we align AI to!

I am thrilled to finally share this work with you all. This is one I am particularly proud of, and it's been an absolute pleasure working with such a talented group of people.

I'm at my first ever Neurips this week! Thanks to everyone who came to our poster yesterday on "Fine-tuning language models to find agreement among humans with diverse preferences" with @bakkermichiel, @MartinJChadwick, @mhtessler and others at @DeepMind #NeurIPS2022

MP confuses Jeremy Hunt as the new Prime Minister

I think it's probably time to share this...I wrote a book. It's coming out in dec 2022 (in theory). It is intended to help cognitive scientists, neuroscientists and AI researchers find a common language for talking about computation and intelligence.

Dramatron paper out! Co-Writing Screenplays and Theatre Scripts with Language Models: An Evaluation by Industry Professionals 📝: https://t.co/i4PqH4m4iI Dramatron is system that enables writers to co-create theatre / film scripts with a language model, @DeepMind's Chinchilla

Large language models can exhibit falsehoods, discriminatory language, and other unsafe behaviour. Introducing Sparrow: a dialogue agent that can search the internet and is trained to be more helpful, correct, and harmless using RL from human feedback: https://t.co/8VtsFjuGcm 1/

Faithful Reasoning Using Large Language Models abs: https://t.co/PKQDgzCdF3

To be more universally effective, efforts to prevent depression might ultimately need to demand less of teachers and kids, and focus more on poverty, violence, homelessness, food insecurity and other structural problems that affect children’s mental health. https://t.co/8UQqzEpiEY

Where did the pandemic begin? Was it from nature or a lab? Since the start, this fundamental question has gone unanswered. Until now. Out in @ScienceMagazine: SARS-CoV-2 emerged into humans via the live animal trade at the Huanan Seafood Market. https://t.co/hnl9j3E6j6

https://t.co/bnSdKzWI32

We have the top tweet of the year right here ladies and gentlemen.

📢 JOB ALERT🚨 Very, *very* excited that we're hiring for a new Ethics Research Associate at DeepMind - join our team of interdisciplinary researchers at the cutting edge of evaluating AI systems and their impact on people. Please share with your networks!

very happy this is finally out!

Two weeks left to apply!