Robyn Speer

@r_speer

Chief scientist at Luminoso. Lead maintainer of ConceptNet. Creator of cutting-edge, knowledge-informed NLP. My pronouns are she/her.

Vrijeme pridruživanja: ožujak 2009.

Tweetovi

Blokirali ste korisnika/cu @r_speer

Jeste li sigurni da želite vidjeti te tweetove? Time nećete deblokirati korisnika/cu @r_speer

  1. Prikvačeni tweet
    28. svi 2019.

    I make ConceptNet and some other things you might find useful in your research. And I am a trans woman. If you use my stuff, please _always_ cite me by my name, Robyn Speer. (See thread for more details)

    Prikaži ovu nit
    Poništi
  2. prije 13 sati

    You already know software is bad. It's even in the news today. But if you needed a reminder of how software is so bad that to make it good we have to fight against _other software_, just told me to google the word "kcah_acitats".

    Poništi
  3. proslijedio/la je Tweet
    3. velj

    There's no excuse for choosing to deadname, least of all from a professional organization.

    Poništi
  4. 3. velj

    But let's back up. What 's leadership is telling me is that they are aware they are deadnaming me and they intend to continue doing so. They claim their hands are tied by their own policies, and this silliness about numbers indicates they don't take the issue seriously.

    Prikaži ovu nit
    Poništi
  5. 3. velj

    (Yes, that's my ORCID. I'm fine with using ORCID for its purpose, but it is not in _any way_ a replacement for the way people use names in research.)

    Prikaži ovu nit
    Poništi
  6. 3. velj

    2. She suggests that it would solve some sort of problem if, instead of by my name, academics were to call me by a number, such as 0000-0003-1682-4533. I asked for clarification and got the same things restated.

    Prikaži ovu nit
    Poništi
  7. 3. velj

    I reached the Associate Director of IEEE, who told me two things: 1. If someone cites me by a man's name, that is an "unfortunate mistake" but it does not need to be remedied in any way.

    Prikaži ovu nit
    Poništi
  8. 30. sij

    Barcelona has been nice

    Poništi
  9. 30. sij

    like I know what they mean and it's not ultimately a big deal, but it's more than just a theoretical quibble when I understand this conference ran out of they/them stickers on the first day

    Prikaži ovu nit
    Poništi
  10. 30. sij

    claiming a "perfect balance of gender" in their review committee, and I just have to continue scratching my head at how gender's being represented here How many non-binary reviewers does that mean there were? How many would represent a "perfect balance"?

    Prikaži ovu nit
    Poništi
  11. 30. sij

    Sometimes you need code or math. Sometimes you need sociology. Sometimes you need UX. Sometimes you need law. Sometimes it's hard to figure out which you need. This conference has done a great job of putting people who do various ones of those in the same room

    Prikaži ovu nit
    Poništi
  12. 30. sij

    and then there are AI applications that are not appropriate, and will not become appropriate with better design, like face recognition. That doesn't need code. It needs regulation to shut it the hell down.

    Prikaži ovu nit
    Poništi
  13. 30. sij

    But if your NLP system is, say, trying to moderate social media, and silencing people when they say the wrong words without regard to context: "don't worry, we de-biased it" is not going to be the answer. It's not going to fix the flawed design

    Prikaži ovu nit
    Poništi
  14. 30. sij

    So you might have seen, I'm a fan of de-biasing as a layer in low-stakes NLP applications. If your system just helps people and computers converse a little better, it's a win when you discourage racism and misogyny in the conversation, so let's apply some de-biasing

    Prikaži ovu nit
    Poništi
  15. 30. sij

    My big takeaway from has been understanding the context of the tension between "algorithmic" and "anti-algorithmic" solutions Sometimes there's a problem where you should do some code or math to make it better Sometimes you just need to shut down the system

    Prikaži ovu nit
    Poništi
  16. 30. sij

    I heard organizers saying "EtherPad" a lot as a way of asking questions I figured it had to be some new thing with that name, not the amazing live text editor of the '00s that Google bought and shut down But it's the text editor! Is it finally back and open and usable?

    Poništi
  17. proslijedio/la je Tweet
    29. sij

    If everyone in the research community is going to keep using COMPAS blindly as the toy dataset for their work, they need to spend some time understanding the criminal justice system better. COMPAS, at least one studied by propublica, is not a bail related model folks

    Prikaži ovu nit
    Poništi
  18. 29. sij
    Prikaži ovu nit
    Poništi
  19. 29. sij

    Pymetrics (the AI hiring company) has been called out by two different talks in this block Especially for sloppy "don't worry, we fixed it!" claims about debiasing Now we have a questioner from Pymetrics asking, what information should we provide to you to help

    Prikaži ovu nit
    Poništi
  20. 29. sij

    Welp it's gone, Web Twitter doesn't understand the idea of losing your connection

    Prikaži ovu nit
    Poništi
  21. 29. sij

    Manish Raghavan talks about evaluating fairness claims in automated hiring products I think this is where de-biasing gets its bad reputation: when companies implementing fundamentally unfair AI say "Don't worry, we fixed it"

    Poništi

Čini se da učitavanje traje već neko vrijeme.

Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.

    Možda bi vam se svidjelo i ovo:

    ·