6 ms·

Machine learning

riffraff 4y ago
But surely ML did advance a lot in recent years? As the saying goes, science advances one funeral at a time, but it does advance. It seems overall you can still have an optimistic point of view of science while accepting it does not work as well as it could.
- ad404b8a372f2b9 4y ago
  That's fair but that's the systemic view of the issue. When you are in the trenches you can't look at the good and throw away the bad. You have to compare your work to the fraudulent work, spend a lot of time reading and trying to reproduce the fraudulent work, cite the fraudulent work, compete with the fraudulent work, publish as fast they do, with results better than they produce. And even the work that's not fraudulent but just bad or rushed. It's a race to the bottom, and how can we ask that people compromise the quality of their work to compete when the only motivation for doing this work is the love of the craft. It's certainly not the money, or the environment, or the work-life balance, or the prestige.
  - epistasis 4y ago
    > When you are in the trenches you can't look at the good and throw away the bad. When in the trenches of science this is literally your job. What seems like the likely hypothesis, what model that explains the data is good or bad. Papers are not textbooks. They are the boiling cauldron from with some bits eventually emerged as fully cooked facts. The only way that I can square others' experience of science with my own is if they take the Discussion sections, which are free-form extrapolation about future directions, and treat them as if they were asserted as truth.
    - ad404b8a372f2b9 4y ago
      None of what you said relates to the problem I described. It relates superficially to the sentence you quoted, but only if you strip it of its context. I'm talking about malicious, fraudulent results. Made up numbers, code obfuscated and manipulated to do things different from what's being claimed. I resent the implication that my issues stem from a naive, idealistic reading of the discussion section rather than a thorough examination of the methods. Sorting the good from the bad is one thing, creating and evaluating models, good models or bad models is one thing. Having to compete AND CITE fraudulent work, lies, imaginary models to be able to participate in the system is not "literally" science. And it's not the job any self-respecting scientist should want to do.
      - epistasis 4y ago
        I'm sorry. The idea that there is outright fraud running rampant is so foreign to me that I clearly can not interpret your experience very well.

Machine learning

But surely ML did advance a lot in recent years? As the saying goes, science advances one funeral at a time, but it does advance. It seems overall you can still have an optimistic point of view of science while accepting it does not work as well as it could.

ad404b8a372f2b9 4y ago
That's fair but that's the systemic view of the issue. When you are in the trenches you can't look at the good and throw away the bad. You have to compare your work to the fraudulent work, spend a lot of time reading and trying to reproduce the fraudulent work, cite the fraudulent work, compete with the fraudulent work, publish as fast they do, with results better than they produce. And even the work that's not fraudulent but just bad or rushed. It's a race to the bottom, and how can we ask that people compromise the quality of their work to compete when the only motivation for doing this work is the love of the craft. It's certainly not the money, or the environment, or the work-life balance, or the prestige.
- epistasis 4y ago
  > When you are in the trenches you can't look at the good and throw away the bad. When in the trenches of science this is literally your job. What seems like the likely hypothesis, what model that explains the data is good or bad. Papers are not textbooks. They are the boiling cauldron from with some bits eventually emerged as fully cooked facts. The only way that I can square others' experience of science with my own is if they take the Discussion sections, which are free-form extrapolation about future directions, and treat them as if they were asserted as truth.
  - ad404b8a372f2b9 4y ago
    None of what you said relates to the problem I described. It relates superficially to the sentence you quoted, but only if you strip it of its context. I'm talking about malicious, fraudulent results. Made up numbers, code obfuscated and manipulated to do things different from what's being claimed. I resent the implication that my issues stem from a naive, idealistic reading of the discussion section rather than a thorough examination of the methods. Sorting the good from the bad is one thing, creating and evaluating models, good models or bad models is one thing. Having to compete AND CITE fraudulent work, lies, imaginary models to be able to participate in the system is not "literally" science. And it's not the job any self-respecting scientist should want to do.
    - epistasis 4y ago
      I'm sorry. The idea that there is outright fraud running rampant is so foreign to me that I clearly can not interpret your experience very well.