Replying to
Here's me wearing jewelry she made, New Years Eve 2016. =)
Remote
mmitchell_ai
@mmitchell_ai@mastodon.social
mastodon 4.8.0-nightly.2026-10-06Interdisciplinary researcher focused on shaping AI towards long-term positive goals. Focusing on ML and ethics.
Opinions mine. =)
0 Followers
0 Following
19 Posts
Joined October 21, 2022
Sorry to everyone hurt by Israel/Palestine attacks. Everyone who has lost people they love, cared for. All the people further hurt by rhetoric online. Death and 280-char tweets don't mix well; the realities here are too horrifying.
I am excited to finally share an idea that has been cooking for awhile: Measuring Data
https://arxiv.org/abs/2212.05129
When you "measure data", you quantify its characteristics to support dataset comparison & curation.
You also begin to know what systems will learn.
I was invited to guest host a MAIHT3000!
Tune in noon PDT, May 20th, for a discussion with me and
@emilymbender@mastodon.social
on facts & fantasy underlying the idea of "AI deception".
http://twitch.tv/dair_institute
https://dair-community.social/@emilymbender/112452295980744218
Replying to
1b: Some of you are familiar with "provenance" information, which is about, "where did this generated thing come from?" This tech now allows *anyone* to do that *VIA AN INVISIBLE WATERMARK* (2/n)
Replying to
2a: Some of you know about how important METADATA is in ethical AI work, to be shared alongside content, disclosing what the content is -- for example, if it's AI generated -- etc. THIS DOES THAT. (3/n)
Replying to
2b: And it does it in the sophisticated "C2PA" compliant way, which roughly means that it follows well-established rules on the right way to do this based on years of work. (4/4)
OH MY GOD. Word on the street is that there's been a 23andMe hack targeting Ashkenazi Jews?! Journalists, can anyone confirm this?!?!
Replying to
1a: Many of you are now familiar with the idea of "invisible watermarks" for generative AI technology. It was one of the White House AI commitments. This tech now allows *anyone* to do it (1/n)
Another paper on model unlearning (YAY), although they missed the opportunity for the title to have OBLIVIATE ✨ ! (H/T
Joshua Lochner)
https://browse.arxiv.org/abs/2310.02238
Is anyone working on AI stuff successfully getting everything they need to do done?
I'm officially holiday-cheered!
Hooray for @evangreer@mastodon.online and Joy for putting this together!!
https://www.instagram.com/reel/C1JAyMCskZX/?igsh=czh4YjVjY3hmMnY4
Replying to
Please lmk if I can contact you with details about an experiment I'm thinking of running, using the form here: https://docs.google.com/forms/d/e/1FAIpQLSfC4MpzQX20cpcJ3TfJIQoSnFjH1czNfZZ99BRlpL5MFwC5-w/viewform
