3 ms·

Journal articles/conference papers are not the only outlet, you can still write technical monographs if you feel review cycles are holding you back.

by jll29 3y ago

godelski 3y ago
It depends. Right now I'm a grad student and I'm just trying to graduate. My friend, who already left, summed it up pretty well. > I don't want to be a successful academic, I want to be a successful scientist. Which I believe are no longer the same thing. I'm just trying graduate and just have to figure out how to play the game enough to leave. Honestly, I do not see myself ever submitting to a journal or conference again. I'll submit to OpenReview, ArXiv, and my blog. I already open my works to discussions on GitHub and am very active in responses and do actually appreciate critiques (there's lots of room for improvement). In fact, my most cited work has been rejected many times, but we also have a well known blog post as a result, and even more than a year later we get questions on our GitHub (which we still respond to, even though many are naive and asks for help debugging python, not our code). But I'm done with academia because I have no more faith in it. I'd love to actually start or work for a truly open ML research group, where it is possible to explore seemingly naive or unpopular ideas, to not just accept things the way they are and forced to chase hype. I will turn down lots of money to do such a thing. To not just metric hack but be honest about limitations of my works and what still needs to be done, that saying such things is not simply giving ammunition to those who would use it against me. To do research that takes time rather than chase a moving goalpost, against people with more compute who rely on pay to play, nor work in this idiotic publish or perish paradigm. To not be beholden to massive compute or to be restricted to only be able to tune what monoliths have been created. To challenge the LLM and Diffusion paradigms that are so woefully incomplete despite there undeniable success. To openly recognize that both these things can be true without it being misinterpreted as undermining these successes. You'd think academia would be the place for this, but I haven't seen a shred of evidence that it is. I've only seen the issues grow.

Journal articles/conference papers are not the only outlet, you can still write technical monographs if you feel review cycles are holding you back.

by jll29 3y ago

Journal articles/conference papers are not the only outlet, you can still write technical monographs if you feel review cycles are holding you back.

godelski 3y ago

It depends. Right now I'm a grad student and I'm just trying to graduate. My friend, who already left, summed it up pretty well. > I don't want to be a successful academic, I want to be a successful scientist. Which I believe are no longer the same thing. I'm just trying graduate and just have to figure out how to play the game enough to leave. Honestly, I do not see myself ever submitting to a journal or conference again. I'll submit to OpenReview, ArXiv, and my blog. I already open my works to discussions on GitHub and am very active in responses and do actually appreciate critiques (there's lots of room for improvement). In fact, my most cited work has been rejected many times, but we also have a well known blog post as a result, and even more than a year later we get questions on our GitHub (which we still respond to, even though many are naive and asks for help debugging python, not our code). But I'm done with academia because I have no more faith in it. I'd love to actually start or work for a truly open ML research group, where it is possible to explore seemingly naive or unpopular ideas, to not just accept things the way they are and forced to chase hype. I will turn down lots of money to do such a thing. To not just metric hack but be honest about limitations of my works and what still needs to be done, that saying such things is not simply giving ammunition to those who would use it against me. To do research that takes time rather than chase a moving goalpost, against people with more compute who rely on pay to play, nor work in this idiotic publish or perish paradigm. To not be beholden to massive compute or to be restricted to only be able to tune what monoliths have been created. To challenge the LLM and Diffusion paradigms that are so woefully incomplete despite there undeniable success. To openly recognize that both these things can be true without it being misinterpreted as undermining these successes. You'd think academia would be the place for this, but I haven't seen a shred of evidence that it is. I've only seen the issues grow.