fun working week
work
This week was particularly fun and productive. I was able to visit Google and Columbia, and I feel like I got a lot of useful work done every day of the week. On Monday, I spend time with my coauthors hammering out some details of our paper and deciding on future research for us to work on. Then, on Thursday, I already had an idea that I'm excited to write-up for that new project. The whole time, I was able to keep up with my meetings and writing for other projects too.
I also put out my first PhD Student Ad! I'm trying to search for students to work with as early as possible, since I won't get to teach the students here at Stevens until the Spring. So, if you know anyone who likes cryptography or CS theory, encourage them to apply!
non-work
Outside of work, I've started reading The Year of the Flood by Margaret Atwood. I'm a little sad that it changed perspective from the first book, but it's still interesting to lean more about the dystopia that Margaret portrays. At this point, I'm invested enough that I'll probably finish the entire MaddAddam series. Then, I'm not sure if there are
I also started a new Baldur's Gate 3 run with a couple of my friends! It's been a while since my first playthrough. This time we're going to try to focus more on the character stories and also play it on the hard mode. It should be fun!
things i saw this week
A YouTube video by Matt Brown outlining some authenticity-related shortcomings of Bitchat
A Boaz Barak blog which is an intro to his AI safety course, but I found the summaries of the speculative scenarios interesting
A MIT Article which illustrates how much personal information is in AI datasets
I learned that almost all printers put tracking dots into the documents they print (!)
Proton launched a privacy-focused LLM with a cat mascot
A substack article about how AI is being forced into products these days
questions
What is the best job scheduling algorithm in terms of approximation ratio
I found out about this on Kunal's website and I've kind of become obsessed with the problem
Can unbounded error-correcting codes be improved? These could be useful for generative AI watermarking (!)
Thanks for reading to the end! Here's a picture from this week:
