NYU professor Tal Linzen submits four papers to NeurIPS
Tal Linzen, Associate Professor of Linguistics and Data Science at NYU and director of the CAP Lab, limited lab size and external collaborations to sustain direct involvement across projects. He submitted four papers to NeurIPS, his highest volume for any single conference, while prioritizing analysis, theory, and model organisms. Leshem Choshen observed that journal publications often support more thorough student work than high-volume conference cycles.
Thanks, that’s kind! A few things off the top of my head:
I’ve never tried to have a big group or start lots of collaborations, which means I can be fairly involved in most projects I work on. I just did four NeurIPS submissions, which I think is the most I have ever submitted to a single conference, and that really maxed out my ability to contribute as much as I would like to. Can’t imagine 10 submissions.
I try to focus on projects with a heavier analysis / theory / “model organisms”, etc component, where there’s less competition and time pressure to publish prematurely, and more competitive advantage for depth and experience.
With my cognitive science work, the goal is scientific. Science journals reward deeper work (most of the reviews are by experts, hard as it may be to imagine!), and there’s less fear of scooping.
Even for pure AI/LLM work I’ve been trying to target more journals recently but that’s not always feasible. Again reward deeper, more complete work.
Leshem (Legend) Choshen 🤖🤗@LChoshen@aliceoh @momergul_ surroundings make a really good job in going deep rather than frequent, maybe @tallinzen group as well, something conscious you do or notice that works?
What kind of benefit do you have in mind? I like the TACL model a lot because you can also present at a conference, so you get the best of both worlds. But also if published in a well known journal the work (and the student) might get more visibility than if it's just one of 5000 NeurIPS papers... See e.g. this paper https://www.nature.com/articles/s41467-025-67998-6
Leshem (Legend) Choshen 🤖🤗@LChoshen@tallinzen @aliceoh @momergul_ Does academic life similarly reward journals? The feeling often is that in this field journals don't benefit the students much more
@LChoshen @aliceoh @momergul_ Sure, yes. Good to submit some work to the big lottery conference sometimes, I do that all the time
Leshem (Legend) Choshen 🤖🤗@LChoshen@tallinzen @aliceoh @momergul_ So visibility is one and the other is that people will care in faculty committees, maybe invited talks or such things that might need a proof a student os successful by the fields standards
@tallinzen @aliceoh @momergul_ Does academic life similarly reward journals? The feeling often is that in this field journals don't benefit the students much more
Tal Linzen@tallinzenThanks, that’s kind! A few things off the top of my head: I’ve never tried to have a big group or start lots of collaborations, which means I can be fairly involved in most projects I work on. I just did four NeurIPS submissions, which I think is the most I have ever submitted to a single conference, and that really maxed out my ability to contribute as much as I would like to. Can’t imagine 10 submissions. I try to focus on projects with a heavier analysis / theory / “model organisms”, etc component, where there’s less competition and time pressure to publish prematurely, and more competitive advantage for depth and experience. With my cognitive science work, the goal is scientific. Science journals reward deeper work (most of the reviews are by experts, hard as it may be to imagine!), and there’s less fear of scooping. Even for pure AI/LLM work I’ve been trying to target more journals recently but that’s not always feasible. Again reward deeper, more complete work.
@tallinzen @aliceoh @momergul_ So visibility is one and the other is that people will care in faculty committees, maybe invited talks or such things that might need a proof a student os successful by the fields standards
Tal Linzen@tallinzenWhat kind of benefit do you have in mind? I like the TACL model a lot because you can also present at a conference, so you get the best of both worlds. But also if published in a well known journal the work (and the student) might get more visibility than if it's just one of 5000 NeurIPS papers... See e.g. this paper https://www.nature.com/articles/s41467-025-67998-6