lukestein’s avatarlukestein’s Twitter Archive—№ 5,979

          1. You want to maximize the probability of selecting a high-quality candidate. High can pass 100% of tests; low can only pass 90%. You can give two (simultaneous) tests. Do you give them to a single candidate, or two different candidates? How do you choose the candidate(s) to test? @Ed_Van_Wesep/1228811770418139136
            Permalink On twitter.com Twitter logo lukestein’s avatar ♻️ 8 Retweets ❤️ 25 Favorites Mood +1 🙂
        1. …in reply to @lukestein
          SPOILER ALERT 🚨 Test one candidate twice. She gets picked 100% if she’s high type, and 81% even if she’s low! How do you pick which candidate to test? Randomly 🤷🏻‍♂️, unless you have even an epsilon bias for/against some candidates. 2/5
          Permalink On twitter.com Twitter logo lukestein’s avatar ♻️ 3 Retweets ❤️ 1 Favorite Mood +1 🙂
      1. …in reply to @lukestein
        .@Ed_Van_Wesep, Shaun, and Brian show that in a much broader class of models (eg, continuous learning, costly learning rather than a fixed budget), it’s NEVER optimal to learn equally about all candidates. Tiebreaking (statistical discrim OR animus) can get hugely amplified. 3/5
        Permalink On twitter.com Twitter logo lukestein’s avatar ♻️ 3 Retweets ❤️ 2 Favorites Mood -2 🙁
    1. …in reply to @lukestein
      Depending on parameters, favored candidates can be screened more intensively (looking for rare disqualifiers) or less intensively (looking for diamonds in the rough 💎). Implications not only for classic “discrimination” questions (eg, gender, race), but also… 4/5
      Permalink On twitter.com Twitter logo lukestein’s avatar ♻️ 3 Retweets ❤️ 2 Favorites Mood +4 🙂
  1. …in reply to @lukestein
    … why do organizations promote internally rather than hire externally? Why do we do in-person interviews at the ASSAs and flyouts rather than Skype? Monogamy? Loved chatting w @Ed_Van_Wesep about their idea last summer, and thrilled they wrote it! 5/5 @lukestein/1228346787519500290 https://t.co/LpQplHOo2B
    On twitter.com Twitter logo lukestein’s avatar ♻️ 3 Retweets ❤️ 6 Favorites Mood +9 🙂
    1. …in reply to @lukestein
      ⚠️Math⚠️ Suppose fraction p are HQ Testing one twice, accept unless we see a failure gives probability of success p + (1-p)(1-0.9²)p Test two, accept a passer (randomize if both pass gives p² + 2p(1-p)((0.9)(½)+(1-0.9)) + (1-p)²(1-0.9)²p The former is bigger for p∈(0,1].■
      oh my god twitter doesn’t include alt text from images in their APIoh my god twitter doesn’t include alt text from images in their API
      Permalink On twitter.com Twitter logo lukestein’s avatar ♻️ 3 Retweets ❤️ 4 Favorites Mood 0