How has publishing your research on LW or X been helpful to you?

Logan Riggs

I work at CBAI, and we might push to get fellows to publish results on LW or X. I'd like to give more than just my own anecdotes.

So please comment your experience here (or in this google form if you don't want a public comment).

My Experience

Better Communication

Arxiv papers take a while to write and aren't optimized for clarity. Blog posts though can be. Once I've done the hard work of writing a clear explanation of my research, then:

Communicating in person is much easier; it's been crytallized/made-a-part-of-me and cached.
The post itself can be linked, or the images & structure re-used for presentations, papers, posters.

Doing this a lot, I've also become a better technical communicator (though this is also accompanied by 30+ min chats where I'm communicating to 1 person at a time. Much easier to get feedback on clarity this way.)

Better People

From LW posts, I've been reached out to for 3 different interp start-ups (bio-interp this, bio-interp that), which isn't my cup of tea.

I have reached out to 10+ folks based off their LW posts ("Hey cool post, I think my work on X could work? Wanna chat; here's my calendly"). On two different occasions:

Messaging Lee Sharkey^[1] about their new SAE work, which led to me co-authoring the SAE paper with Lee and his MATS mentee Hoagy.
Commenting on Thomas Doom's post on tokenized-SAEs, led to reaching out to him a year later about weight-based interp. We now work on tensor networks together (which are performant & let you do cool interp with)!

Papa Johns Better Feedback

I honestly don't think I've recieved great feedback on my research through my LW posts (definitely interesting ideas, but nothing game-changing). Most of the good ideas come from discussing it with folks in 30+ min calls or in person chats.

I do think I've left good advice on other's post (only possibly gamechanging lol), and benefited personally from engaging with their work.

I'm very interested in hearing other's experiences though; especially:

Experiences much different than mine!
Any experience on twitter/X (I just don't get on ~~sadly~~ happily)
Tips and tricks you've learned

[And again, the overall purpose is for advice to mentees through a research fellowship program]

^{^}
I did know Lee Sharkey a bit from AI safety camp, virtually, with most of our talking over 1 catch-up call from after the program. I think Lee would've responded anyways though; he's a nice guy.

what kind of research are you interested in? Just AI?

My rationale is that the point of research, unless you work at a lab, is that someone will read it and do something with it, if you publish to LW or X you would get faster feedback than on arxiv, where someone would only read it if they were doing a literature review or reading your resume. I imagine posting on arxiv works if you only care about references or building a resume, but for AI safety where speed matters waiting for citations is very slow. On LW I haven't got technical advice but got questions and DMs about how someone used my research. And on X (I didn't post but a collaborator did) we got technical advice. But the point is that even just a little indirect feedback is better than nothing.

Also another thing is that on papers one might be incentivized to only write technical results and leave out the informal "intuitions" which might end up being important, while LW has more of a "rubber duck debugging" effect.