Imposter Syndrome, Stupid Questions, and Eight (+/-2) Problems I Need your Help With
summary
Claims that many ML papers have algorithm and implementation mismatches and that bugs can make broken methods appear to work; demonstrates an optimal sparse decision-tree solver using dynamic programming with tight pruning bounds, prototyped in Python then rewritten in C++ (and ported toward Daphne), finding notation and proof bugs along the way. Builds synthesis tools that generate VM/OS modules (a “dancing bear” that runs) and a template‑component C++ kernel assembler for tiny kernels (~80–120 LOC). Asks for tools to link proofs, generate test datasets, and specify/hierarchically synthesize ISAs and kernels.
AI-written summary · about
chapters (19)
- 0:00Start
- 0:51Systems - Construed Broadly
- 2:01Collaboration
- 3:06My Other Team
- 5:23Once Upon a Time
- 6:24The Villain
- 8:31The Hero: The Stupid Question
- 9:26The Happily Ever After
- 11:36The Problem with ML ...
- 17:56The Development Process
- 19:19One Approach
- 24:28Late-breaking Related Problem
- 25:51From Operating Systems to Program Synthesis
- 29:26What did we Learn?
- 32:54BOS Module Architecture
- 33:25Balancing Flexibility and Performance
- 34:04Fun questions in Program Synthesis
- 35:21Variants on Program Synthesis
- 38:08My Team
related talks
22mWhat Developers and Economists Can Learn from Each Other!
05m10 Things Junior C++ Devs Don't Get
19mAI missteps as stepping stones
02h 04m6 impossible things
56mThings you thought you didn’t need to care about that have a big impact on your job
51mWe're Doing It All Wrong
44mGrasping Complexity with Both Hands
01h 43mCrazy Code, Crazy Coders