Back to the catalog
performancestatisticsinternet-culture

When HBO's Silicon Valley became real: Not Hotdog, the Weissman score and tabs versus spaces

14 questions · by Quizbun

HBO's Silicon Valley hired tech consultants so its jokes would hold up, and a few of those jokes escaped into the real world: a compression metric a Stanford professor invented for the show, a hotdog-detecting app you could actually install, and a fake Google that the real Google linked to. Each Question starts from a scene and asks what happened next in real life. The Explanations cover the real tech behind the jokes: transfer learning, class imbalance, and why a score built on logarithms depends on your stopwatch.

Start

Questions

  1. Not answered. Pied Piper's compression is judged by its "Weissman score". Where did that metric come from?
  2. Not answered. Compute the Weissman score for this compressor.
  3. Not answered. You score the same two compressors with times in minutes, then in seconds, and get two different Weissman scores. What does that tell you?
  4. Not answered. In 2016, two Dropbox engineers reported a real Weissman score in a post on the company's tech blog. What had they built?
  5. Not answered. In August 2015, Google announced its new parent company, Alphabet, on the website abc.xyz. What was hidden in the announcement?
  6. Not answered. In "Bachmanity Insanity" (season 3, episode 6, 2016), Richard breaks up with Winnie, a developer from Facebook. What does he refuse to put up with?
  7. Not answered. In "Teambuilding Exercise" (season 4, episode 4, 2017), Jian-Yang demos SeeFood, an app meant to recognize any food. It can only say "hotdog" or "not hotdog". What happened in the real world?
  8. Not answered. Who built the real Not Hotdog app?
  9. Not answered. Anglade's first version sent photos to a cloud vision API. Which reasons did he give for moving the neural network onto the phone?
  10. Not answered. Anglade started with only a few thousand hotdog photos, far too few to teach a network to see from nothing. How did he get a working first model?
  11. Not answered. The Inception-based model worked, but it didn't ship in the app. Why not?
  12. Not answered. One of Anglade's smaller models seemed to decide "red sauce means hotdog". What went wrong?
  13. Not answered. The final training set had 150,000 photos, of which 3,000 were hotdogs. A lazy model answers "not hotdog" for every photo. What accuracy does it score on this set, in percent?
  14. Not answered. Which of these from Silicon Valley exist, or happened, in the real world?