Kyle Vedder

Vocabulary learned from Kyle:

• Busted
• Chopped
• Clapped
• Cooked

Technical vocabulary that he uses that I want to use more on a day-to-day:

• Surrogate
  • “Honestly, I think the survey data is just a surrogate for what people actually feel — we should watch how they behave instead.”
• Posterior / Prior
  • “your prior on that restaurant was biased by TikTok hype”
  • In Bayesian models: a prior is explicitly defined, e.g., a Gaussian over weights.
  • In everyday ML work: your “prior” is often implicit — baked into your architecture, inductive biases, or data preprocessing.
• Manifold
  • lower-dimensional structure the data lies on
  • There’s hidden structure or simplicity beneath surface complexity. That’s why they have a picture of the curvature
• Ablation - that’s more ian
• Out-of-distribution
• Conditional likelihood
• marginal likelihood
• Mode collapse
  • “Every time we go out, we end up at the same ramen place. Total mode collapse.””