5 ms·

Why the "WTF!"? He wasn't sanctioned for using an LLM to speed up his work, so seeing this as the first reply could confuse some people. I've been doing this

Why the "WTF!"? He wasn't sanctioned for using an LLM to speed up his work, so seeing this as the first reply could confuse some people. I've been doing this for months now too. I can get docs done in a day that used to take a couple weeks because of constant interruptions.

belter 4y ago
That is not the point...It's the fact his supposed valuable research, ( he is getting published after all!) is of so flimsy quality, that a LLM model can be a productivity improvement! That is the reason of the WTF. Do you think Tesla or Einstein with a LLM model, would spit out 5 new discoveries per week instead of one every decade?
- coldtea 4y ago
  >is of so flimsy quality, that a LLM model can be a productivity improvement! Who said the research was of flimsy quality? Who said that the discoveries (as opposed to the BS padding journals require) was what LLM was used for? They might or might not be. But not because he used an LLM to "polish them".
  - belter 4y ago
    Has been publishing sometimes every 5 days, it seems mostly, by being listed as co-author. I like my genius with some effort and sweat. Even Terence Tao talks about his challenges... https://orcid.org/0000-0003-4190-1916 https://orcid.org/0000-0003-4190-1916
    - shadowgovt 4y ago
      What is the nature of the space he's working in? Something like physics, you can have a gap of months or years between a new theoretical breakthrough or the completion of a new machine giving you new raw data to churn through. ... but chemistry is something (no offense to chemists) you can do to the dirt outside your house, and every novel molecular arrangement can be paper-worthy.
      - belter 4y ago
        Let's not go that way... - https://xkcd.com/435/ https://xkcd.com/435/
- rscho 4y ago
  Welcome to capitalistic science, where moar is always better. You know, 40y ago when you bought a tool made of stainless steel that would last for life, and now you get the same out of flimsy plastic? Same thing with science.
- greatpatton 4y ago
  For a non native English speaker, LLMs are really useful to rewrite your texts in a more natural way.
  - whythre 4y ago
    Dude was publishing every 37 hours. He wasn’t just using it to ‘polish’ his language skills, he was churning out absurdly low grade schlock with the aid of an AI and getting it rubber stamped by the broken journal cartel.
    - lesuorac 4y ago
      I mean the guy was writing a paper every 2-3d without AI and got it down to 1d with AI. It seems completely reasonable to me that better tooling could get their time down 50% (try shoveling a driveway vs using a truck ...). If his was just (ab)using chatgpt to create the papers from thin air then it'd be pretty easy for somebody to show where the pre-chatgpt papers are and the post-chatgpt papers are by the drop in quality. However, nobody has been citing the content of his papers when they make the claims about quality so I doubt chatgpt has an impact on the quality of his papers.
    - dgellow 4y ago
      He was “publishing” every 37h, it seems to be mostly by co-signing papers
- twelve40 4y ago
  not a bit disturbed by whatever language tools he needs to use but a paper every day (or every week) is insane. No matter what you do, it can't be sustainable for years as he claims.
  - igetspam 4y ago
    And that is absolutely worthy of discussion but the use of tools is a distraction from the topic. I find it surprising that the author felt it was even worth bringing up in the article. It makes no argument for or against the subject of the article but definitely plants some weird seeds of distrust for certain types of people.
- shadowgovt 4y ago
  This isn't indicative of the quality of his work; it's indicative that the standardized form of the research paper has so much low-information boilerplate that it's semi-automatable. If anything, I applaud attempts to automate it (only slightly less than I applaud attempts to do away with it via a new structure for these papers). Complaining about this is like complaining about the fact that people use IDEs to manage the imports in their Java class files. Ideally, you'd stop using a crap language that makes you wrap three lines of actual implementation in 50 lines of imports, declarations, and redundant-to-the-code comments to make doxygen (and a linter) happy, but if the institutional model prevents that... By all means use a good IDE to do half the work for you.
- adastra22 4y ago
  Have you tried using a LLM? I don’t think they are as dumb as you are making out. I’ve used GitHub Copilot to help with some lab reports this last semester, and it really sped things up. It can auto fill complicated LaTeX syntax and such so that I don’t have to look it up. The meat of the paper is mine, and I have to correct it a lot, but it was definitely useful.

belter 4y ago

That is not the point...It's the fact his supposed valuable research, ( he is getting published after all!) is of so flimsy quality, that a LLM model can be a productivity improvement! That is the reason of the WTF. Do you think Tesla or Einstein with a LLM model, would spit out 5 new discoveries per week instead of one every decade?

coldtea 4y ago
>is of so flimsy quality, that a LLM model can be a productivity improvement! Who said the research was of flimsy quality? Who said that the discoveries (as opposed to the BS padding journals require) was what LLM was used for? They might or might not be. But not because he used an LLM to "polish them".
- belter 4y ago
  Has been publishing sometimes every 5 days, it seems mostly, by being listed as co-author. I like my genius with some effort and sweat. Even Terence Tao talks about his challenges... https://orcid.org/0000-0003-4190-1916 https://orcid.org/0000-0003-4190-1916
  - shadowgovt 4y ago
    What is the nature of the space he's working in? Something like physics, you can have a gap of months or years between a new theoretical breakthrough or the completion of a new machine giving you new raw data to churn through. ... but chemistry is something (no offense to chemists) you can do to the dirt outside your house, and every novel molecular arrangement can be paper-worthy.
    - belter 4y ago
      Let's not go that way... - https://xkcd.com/435/ https://xkcd.com/435/
rscho 4y ago
Welcome to capitalistic science, where moar is always better. You know, 40y ago when you bought a tool made of stainless steel that would last for life, and now you get the same out of flimsy plastic? Same thing with science.
greatpatton 4y ago
For a non native English speaker, LLMs are really useful to rewrite your texts in a more natural way.
- whythre 4y ago
  Dude was publishing every 37 hours. He wasn’t just using it to ‘polish’ his language skills, he was churning out absurdly low grade schlock with the aid of an AI and getting it rubber stamped by the broken journal cartel.
  - lesuorac 4y ago
    I mean the guy was writing a paper every 2-3d without AI and got it down to 1d with AI. It seems completely reasonable to me that better tooling could get their time down 50% (try shoveling a driveway vs using a truck ...). If his was just (ab)using chatgpt to create the papers from thin air then it'd be pretty easy for somebody to show where the pre-chatgpt papers are and the post-chatgpt papers are by the drop in quality. However, nobody has been citing the content of his papers when they make the claims about quality so I doubt chatgpt has an impact on the quality of his papers.
  - dgellow 4y ago
    He was “publishing” every 37h, it seems to be mostly by co-signing papers
twelve40 4y ago
not a bit disturbed by whatever language tools he needs to use but a paper every day (or every week) is insane. No matter what you do, it can't be sustainable for years as he claims.
- igetspam 4y ago
  And that is absolutely worthy of discussion but the use of tools is a distraction from the topic. I find it surprising that the author felt it was even worth bringing up in the article. It makes no argument for or against the subject of the article but definitely plants some weird seeds of distrust for certain types of people.
shadowgovt 4y ago
This isn't indicative of the quality of his work; it's indicative that the standardized form of the research paper has so much low-information boilerplate that it's semi-automatable. If anything, I applaud attempts to automate it (only slightly less than I applaud attempts to do away with it via a new structure for these papers). Complaining about this is like complaining about the fact that people use IDEs to manage the imports in their Java class files. Ideally, you'd stop using a crap language that makes you wrap three lines of actual implementation in 50 lines of imports, declarations, and redundant-to-the-code comments to make doxygen (and a linter) happy, but if the institutional model prevents that... By all means use a good IDE to do half the work for you.
adastra22 4y ago
Have you tried using a LLM? I don’t think they are as dumb as you are making out. I’ve used GitHub Copilot to help with some lab reports this last semester, and it really sped things up. It can auto fill complicated LaTeX syntax and such so that I don’t have to look it up. The meat of the paper is mine, and I have to correct it a lot, but it was definitely useful.