4 ms·

I mean take a look at the names. DoNotAmplify, CoordinatedHarmfulActivityHighRecall, UntrustedUrl, MisleadingHighRecall, NsfwHighPrecision, NsfwHighRecall, Civ

by mhoad 4y ago

I mean take a look at the names. DoNotAmplify, CoordinatedHarmfulActivityHighRecall, UntrustedUrl, MisleadingHighRecall, NsfwHighPrecision, NsfwHighRecall, CivicIntegrityMisinfo, MedicalMisinfo, GenericMisinfo, DmcaWithheld, HatefulHighRecall, ViolenceHighRecall, HighToxicityModelScore, UkraineCrisisTopic, DoNotPublicPublish What would you suppose the list is for? Almost everything on that list is something they have publicly stated they are trying to reduce. I don't think we have to play dumb here and say "it didn't specifically use the word downvote therefore we can't possibly know what it is for"

ad404b8a372f2b9 4y ago
This is reverse engineering, not divination. I just gave you a number of other algorithmic functions that would be just as likely to use such a list.
- oneeyedpigeon 4y ago
  > you can make some educated guesses based on naming This is what people are doing. > but generally you want to check those guesses by seeing how, where and why a variable is used Agreed, but without the full source code available, this simply isn't possible.
  - ad404b8a372f2b9 4y ago
    That's fair which is why I'm asking if a piece of code has been published supporting the claim of down-ranking. If it hasn't the answer is to very carefully qualify the statement "Analysis of Twitter algorithm reveals Twitter down-ranks tweets about Ukraine" rather than treating the people who question it as heretics.
    - oneeyedpigeon 4y ago
      I agree. It shows the danger of selectively releasing code and how the "Twitter open-sources algorithm" rhetoric is deeply misleading, at best. Hopefully, any misplaced analysis of the code will result in greater transparency.
      - ad404b8a372f2b9 4y ago
        Yeah it reminds me of everyone scrambling to draw conclusions from the leaked Yandex SEO algorithm, even though all the models were missing...
  - 93po 4y ago
    > you can make some educated guesses based on naming > This is what people are doing. no, they're reporting on it as fact

by mhoad 4y ago

ad404b8a372f2b9 4y ago

This is reverse engineering, not divination. I just gave you a number of other algorithmic functions that would be just as likely to use such a list.

oneeyedpigeon 4y ago
> you can make some educated guesses based on naming This is what people are doing. > but generally you want to check those guesses by seeing how, where and why a variable is used Agreed, but without the full source code available, this simply isn't possible.
- ad404b8a372f2b9 4y ago
  That's fair which is why I'm asking if a piece of code has been published supporting the claim of down-ranking. If it hasn't the answer is to very carefully qualify the statement "Analysis of Twitter algorithm reveals Twitter down-ranks tweets about Ukraine" rather than treating the people who question it as heretics.
  - oneeyedpigeon 4y ago
    I agree. It shows the danger of selectively releasing code and how the "Twitter open-sources algorithm" rhetoric is deeply misleading, at best. Hopefully, any misplaced analysis of the code will result in greater transparency.
    - ad404b8a372f2b9 4y ago
      Yeah it reminds me of everyone scrambling to draw conclusions from the leaked Yandex SEO algorithm, even though all the models were missing...
- 93po 4y ago
  > you can make some educated guesses based on naming > This is what people are doing. no, they're reporting on it as fact