I wrote this for Slate a couple years ago and find it particularly relevant today in light of all the recent OpenAI drama: slate.com/technology/202…

😊👋
theinnovator.news/interview-of-t…

1/ Today the UK's AI Safety Institute is open sourcing our safety evaluations platform. We call it 'Inspect': gov.uk/government/new…

our story on UAE & Silicon Valley is on the front page of today’s The Washington Post

“Anyone seeing AI companies team up with dictatorships is right to be concerned,” El Mahdi El Mhamdi | #FreeKoulchi
“If you enjoyed WhatsApp images to send spyware, you will love AI models and AI data sets to send

I was invited to guest host a MAIHT3000!
Tune in noon PDT, May 20th, for a discussion with me and @[email protected] on Mastodon on facts & fantasy underlying the idea of 'AI deception'.
twitch.tv/dair_institute

a very nice quote from Kate Crawford at #chi2024 --
'…the loudest alarmists about existential risk are also the architects of the technology they warn against. By sustaining this paradox, they are reaping unprecedented financial investment and political power.'

Meanwhile, Kyle Wiggers low key doing the Lord's work over here. =)

The women in AI making a difference ift.tt/jA9DkOK

Very interesting release that confirms once more that data quality is the new training law: only 500b 'high-quality' tokens (and, from what I can gather, 7 epochs !!)

#TransRightsAreHumanRights

Dr. Margaret Mitchell (MMitchell), a renowned leader in the field of artificial intelligence and ethics, emphasises the ethical considerations in AI development, ensuring technology is used responsibly and justly.

Book the leading expert in AI: londonspeakerbureau.com/speaker-profil…

I love this work from Adji Bousso Dieng and colleagues so much.

I just arrived in San Francisco and I’m starting to review the applications for our SF community lead. Lots of conventional ones - where are the crazy innovative out-of-the-box applications like HF spaces or viral tweets to get my (and the community’s) attention?

Niloofar has been doing very cool work on privacy in language models for a while. If you're interested in the topic I strongly recommend dropping by her posters.

'model behaviour is not determined by architecture, hyperparameters, or optimizer choices. It's determined by your data, nothing else. Everything else is a means to an end in efficiency delivery compute to approximating that data.' reddit.com/r/MachineLearn…

Ali Schwanke When faced with moral questions like this you should ask yourself “does this opinion seek to uplift or does it contribute to someone’s oppression / suffering” and you will have your answer

My husband found it very helpful when he was going through a very long period of job hunting. I don’t see what’s desperate about signaling what you’re looking for on a site that is built exactly for such purposes.

theonion.com/these-kids-sho…

University land acknowledgements being put to the test.

Can't really think of a greater betrayal than calling militarized police on your students for doing their best in horryfing global circumstances.