Omar Khattab, Assistant Professor at MIT CSAIL, observes that frontier AI labs closely track academic machine learning releases shared on social media platforms

this is not some complex inference; you can tell just because they always reach out. but it just conflicts with the facade of mystery and secret knowledge that some like to ascribe to the labs lol.

most recently about pedagogical RL by @NoahZiems* @souradipchakr18*

https://noahziems.com/pedagogical-rl

Most people in this part of twitter don’t realize how closely folks at the frontier labs pay attention to all your favorite academic ML releases on here.

@lateinteraction It’s an annoying one-way relationship, especially with some labs.

@lateinteraction I was very surprised how many/the type of reachouts I started to get when I started building in public. I think even psychologically most people passionate about the field just want friends who share that same energy.

@lateinteraction Just wanna say thank you. This is very empowering.

All the behind the door mystery of labs make you feel you can't make any meaningful contribution if you are in without access to the secrets and the compute.

But more and more you realize that is just not true!

@lateinteraction Industry is very good at scaling things but it lacks humanpower, resources, and willingness to try a gazillion of often risky ideas.

@lateinteraction Very true. Unfortunately it is now unidirectional.

@lateinteraction If I was in a frontier lab I wouldn't take as many risk on my experiments as I do in Academia.

@lateinteraction the lurk list runs deeper than the like count suggests

@lateinteraction feels like a whole surveillance operation behind the timeline

theyre watching us argue about batch size the whole time

@lateinteraction I was just thinking about this, holy shit. I had a feeling their are people @ frontier labs who just stalk papers all the time

@AlexRoseJo Looking forward to seeing you next week!

@lateinteraction fact

@lateinteraction It’s not just frontier labs, it’s also smaller labs and infra people (ok, maybe just me).

@lateinteraction it is the true lifeblood

@lateinteraction academic ML twitter is the only seminar series where attendance is mandatory for the labs but optional for the speaker

@lateinteraction labs engineers are probably the biggest lurkers in academic ML twitter

lowkey validating when they cite a thread in a paper

(not a complex inference; you can tell just because they always reach out!)

@SamadSyed @lateinteraction the reachouts are the best part honestly. had more useful conversations from random DMs after sharing what i was working on than from months of networking. what kind of reachouts surprised you most?