Just like we have a golden rule for crypto, we should have one for generative models too: "Do *not* use a generative model as a search engine"

Twitter is full of those "omg I've run a vm inside chatgpt!" posts these days. The model is so good that fake nvidia drivers are always broken exactly like IRL nvidia drivers

What I thought I wanted: a tool that automatically flags text generated by chatgpt Waht I actually want: a tool that automatically flags fake screenshots of chatgpt doing weird stuff on twitter

working with networkx feels like:

next startup idea: dynamic analysis sandboxes for language models :D

Thank you AK for the retweet! Minimizing the number of edge crossings in graph drawings is an exciting open data vis problem. With Giorgio Victor Schetinger and Cody Dunne, we wondered if it is possible to perform crossing reduction using solely natural language.

CONGRATULATIONS DR. Di BARTOLOMEO! 🥳🎉 We had a blast celebrating Sara Di Bartolomeo 's successful thesis defense Take a look at some of Sara's incredible graph visualization work here: vis.khoury.northeastern.edu/people/Sara-Di… and congrats also to Cody Dunne for being a wonderful advisor 💙

tired: boring online petition with dubious signatures wired: GDPR them back into the stone age

Sorry I had to

I'm pretty sure the latex compiler is already slightly conscious -- and it hates us

When analyzing ML security and privacy you need to study 𝐬𝐲𝐬𝐭𝐞𝐦𝐬, not just models! Our new paper shows that privacy is way worse when models are deployed in systems that use data cleaners, output filters, etc. Paper: arxiv.org/abs/2309.05610 Blog: spylab.ai/blog/side-chan…

This also answers the question: how's your phd going?

Slides for our ACSAC paper "Poisoning Network Flow Classifiers" are online at: acsac.org/2023/files/web… Paper link: dl.acm.org/doi/10.1145/36…