A history correction: I am not the first to discover prompt injection. I was merely the first to do so and discuss it publicly. PI was discovered independently by multiple teams. The first was Preamble, an LLM security company, whose find predates mine by several months.
RESEARCH
-
Curriculum Learning and Dropout Improve Model Performance
By
–
Thanks to using a form of curriculum learning (5 epochs) and better dropout (20 & 50 epochs) I've now been knocked off the leaderboard entirely! Here's the latest:
-
LLM Tokenization: Not Vectors or Letters, but Token Sequences
By
–
It doesn’t see words as embedded vectors or as letters, it sees them as token sequences — chunks of about 4 letters from a fixed mapping of strings to tokens. It does better on letter-based tasks if you ask it to first rewrite words l-i-k-e t-h-i-s.
-
Paper Review Practices at Major AI Conferences
By
–
Yes, that is probably true. I put out my numbers publicly to give at least n = 1 data point. I would be curious to know how many papers are reviewed by those who submit 30+ papers to a single conference. Though maybe their large group would make up any deficit.
-
NASA STEM STEAM Education: 8 Ways Students Explore
By
–
Break Down Boundaries! 8 Ways #Students Can Dive Into #STEM & #STEAM with #NASA in 2023 @NASA https://
bit.ly/NasaSTEAM #education @psb_dc @helene_wpli #Webb #dataScientist @TeachersPLN #Space #AI #ML @MatthewSekol @DigitalColmer @Paula_Piccard @michaeldacosta @IamNankpak -
Balancing Peer Review Contributions in Academic Publishing
By
–
A rough count gave about 14. Each of these papers has at least 3 authors (most have more), and each got ~3 reviews, meaning that to remain review neutral, I'd need to do ~14 reviews.
-
Top 2023 Tech Trends Predicted By AI Model
By
–
Here Are The Top 2023 Trends As Predicted By An #AI Model : A Perfect Storm of Tech https://
medium.datadriveninvestor.com/here-are-the-t
op-2023-trends-as-predicted-by-artificial-intelligence-828a093a0ca6
… @pierrepinna @Xbond49 @PawlowskiMario @psb_dc @gvalan @TheRudinGroup @HaroldSinnott @mikeflache @Shi4Tech @Nicochan33 #MachineLearning #DeepLearning #Fintech #Datascience -
2022 Conference Review Activity Across Seven ML Programs
By
–
2022's year in reviews: I did 28 conference reviews, plus 42 meta-reviews, on 7 program committees (COLT, UAI, NeurIPS, SaTML, ICLR, USENIX Security, ALT). I also chaired two #ICML2022 workshops (TPDP and UpML). Numbers are down from last year, so why do I feel more tired…
-
Differential Privacy and Secure Aggregation in Federated Learning
By
–
It's a bit complicated. Methods like differential privacy and secure aggregation on top of FL generally help, but there are some caveats. See "Is it possible to Prevent Our Passive and Active Attacks?" of http://
cleverhans.io/2022/04/17/fl-
privacy.html
…. -
Federated Learning: Heterogeneity and Privacy Perspectives
By
–
I like FL as a lens through which to study heterogeneity in clients, which may have different distributions, resources, or capabilities. But not for privacy. Here is another perspective on privacy of FL, which is a bit more conspiratorial than my own https://
x.com/le_science4all
/status/1602432680657928193
…