Tesla FSD insider: end-to-end is settled, winning hinges on data diet quality
Singularitarian · x · 2026-11-23
Tesla FSD team member yuntatsai responded to whether Elon could just write down the 'magic' for v15: anyone can write the plan; what matters is execution speed and care.
He notes many ML researchers already agree end-to-end is the way, but you can no longer spoon-feed junk data — even with the right algorithm and architecture. The team's effort ensuring the model gets the right 'diet and nutrition' is insane, likening ML training to running a Michelin-starred kitchen from executive chef to prep. The thread also references Bill Freeman's CVPR 2026 talk on bitter lessons.
More from Companies & People
- Microsoft offers voluntary exits to ~60,000 employees as AI-native orgs upend incumbents — thedealdirector · 2026-11-23
- Granola Cofounder on 'Fishing for a Problem' and the Design Behind Its Viral Onboarding — soleio · 2026-11-23
- Tobi Lütke on Jensen: Nvidia's product 'takes rocks and makes them think' — MartinGTobias · 2026-11-23
- Why OpenAI is efficiency-obsessed: serving 1.2B users could bankrupt it — VraserX · 2026-11-23
- Gurman: Apple isn't organized to succeed in the AI era, needs overhaul — jdluk87 · 2026-11-23
- AI Alliance, advised by LeCun, launches Project Tapestry for federated frontier model training — anshulkundaje · 2026-11-23