Last week @Trevornoah asked @OpenAI @miramurati
: How can we safeguard against AI-powered photo editing for misinformation? MIT students hacked a way to "immunize" photos against edits: http://
gradientscience.org/photoguard/ @aleks_madry
RESEARCH
-
MIT immunizes photos against AI-powered misinformation edits
By
–
-
D&D Alignment Chart Maps Nine Ethical Lying Preferences
By
–
Interestingly, the nine preferences that describe the conditions when people feel it okay to lie fit surprisingly well in the classic D&D alignment chart. There is even a true neutral! https://
katharinajanezic.com/static/Janezic
_JMP.pdf
… -
Elite Knowledge Workers Drove Industrial Revolution Growth Not Mass Education
By
–
A puzzle about the Industrial Revolution: unlike now, there seemed to be little correlation between the education of workers & early industrial growth. This paper shows why: it was the presence of highly educated elite knowledge workers (& their inventions) that powered growth.
-
Reincarnating RL: Efficient Reinforcement Learning from Prior Computation
By
–
Training #ReinforcementLearning algorithms from scratch is computationally intensive and time consuming. We propose an alternate approach, Reincarnating RL, that integrates prior computation into the RL training workflow. Learn more and grab the code at https://t.co/XwfX0uakJZ pic.twitter.com/ktaQznCD1a
— Google AI (@GoogleAI) 3 novembre 2022Training #ReinforcementLearning algorithms from scratch is computationally intensive and time consuming. We propose an alternate approach, Reincarnating RL, that integrates prior computation into the RL training workflow. Learn more and grab the code at https://
goo.gle/3Ws2TLk -
Unethical AC Review Practices: Weaponizing Citations Against Researchers
By
–
Unethical pro-tip: if you're an AC and you want to make someone else look bad, write a terrible meta-review and demand citations to many of their papers. Reminder that, despite the repeated authors here, we don't know who the ACs are. Don't jump to conclusions.
-
Bay Area Data Science Community ML Meetup Fireside Chat
By
–
Bay Area Data Science Community: Don’t miss our first #ML Meetup! Join us next Tues at 5pm for food, drinks, and an engaging fireside chat with @DalianaLiu and @w4nderlus7
, CEO of Predibase and founder of @ludwig_ai
. Save your spot: -
Appliance Mode for Training Massive ML Models Introduced
By
–
ICYMI we recently had a developer community meeting where we introduced our Appliance Mode for training massive ML models. You can watch our recording here: https://
hubs.li/Q01rkJ9S0 Want to join our next meeting? Sign up here! https://
hubs.li/Q01rkHTb0 #ML #machinelearning #ai -
Runway’s Journey: From NYU Thesis to AI Leadership
By
–
I've had a front-row seat to Runway's evolution from the time it was just @c_valenzuelab's thesis project at NYU in early 2018. Very few people had the foresight back then to start building towards something like this. Super proud of the whole team for where they are today. https://t.co/cbJihnSo7V
— Gene Kogan (@genekogan) 3 novembre 2022I've had a front-row seat to Runway's evolution from the time it was just @c_valenzuelab
's thesis project at NYU in early 2018. Very few people had the foresight back then to start building towards something like this. Super proud of the whole team for where they are today. -
Empathetic peer review in AI research publication process
By
–
Usually, we think of AC rejections and being cold and uncaring. But this AC was quite empathetic, picking up on a self-critical comment from the author and supporting them in their journey. https://
openreview.net/forum?id=A6O79
ipjlJC
… -
Paper Rejection Despite Acceptance Scores Highlights Review Process Issues
By
–
The paper was rejected despite all scores being "Accept" or better, because the AC says they didn't compare with some other papers, which were never mentioned before. The AC should have raised this to the authors earlier, or figured out the relevance with help of the reviewers.