The profession of solving (the wrong problem)

on YouTube ↗35m

" summary

Concrete outcomes and methods: a classroom regression gives an A+ but misses capacity limits; a recommender stack built with a database, cron/Airflow and A/A test rejects DeepLearning for a simple next-episode heuristic that immediately boosts engagement; a Kaggle challenge is broken by brute-force seed discovery (≈4 billion seeds); food logistics improves by rephrase beans→nutrients (choose lentils nearby instead of solving NP‑hard routing); a timeseries scheduling problem gets solved by ignoring the signal and changing production to a negatively correlated product.

AI-written summary · about

" chapters (28)

1. 0:00Welcome
2. 0:10The Profession of Solving (the Wrong Problem)
3. 0:40Let's start with a true story
4. 1:00Introduction - Assignment for a first year course in statistics
5. 1:25Data from a gig at a local theatre
6. 2:23Statistically significant result
7. 3:04Realizing what went wrong
8. 5:00First story - Designing a recommender system for the NPO
9. 7:00Step 1: DrawPictures not DeepLearning
10. 9:00Step 2: A/A test this setup!
11. 9:51Step 3: alg(B) = entropy
12. 10:41Step 4: More Measuring: the order!
13. 11:37Step 5: Is it time for DeepLearning?
14. 11:55Step 5: not DeepLearning, but DumbThinking
15. 13:13Vincent's Lemma: Serendipity (The solution)
16. 16:14Gotta Love Heuristics: Lessons
17. 17:45Second story - 261 days: a billboard lottery
18. 18:59Mathematical formulation of the billboard problem
19. 20:21Rephrasing the billboard problem
20. 21:21Applying rephrasing to a WHO food problem
21. 25:08Third story - Kaggle's Santa Problem
22. 26:52Mathematical formulation of the problem
23. 27:56Everyone was solving the wrong problem
24. 29:42Lessons learnt
25. 30:55Last story: the condom factory
26. 32:26Solution to the condom factory problem
27. 33:30Parting words
28. 34:40Advice

" related talks

• 24mThe Rise of Reasoning ModelsValentina Halasi · LambdaConf 2025
• 40mPuzzles, Problems, and ProgramsChris Martens · Strange Loop 2018
• 44mAI Engineering Anti-patternsChip Huyen · Craft Conference 2025
• 52mLearning as a Tool for Algorithm Design and Beyond-Worst-Case AnalysisKevin Leyton-Brown · Learning, Algorithm Design and Beyond Worst-Case Analysis
• 19mPractical problem solving for wicked problemsStacy Rendall · FOSS4G SotM Oceania 2023
• 11mTransforming Problem-Solving with Data-Driven InsightsNikhil Bharadwaj Ramashasthri · Conf42 Cloud Native 2025
• 39mUnlocking reasoning in Large language modelsLogesh Kumar Umapathi · Conf42 Machine Learning 2023
• 23mSkills for debugging complex problemsNick Humrich · UtahJS Conf 2023