BG2Pod with Brad Gerstner and Bill Gurley Ep4. Tesla FSD 12, Imitation Learning Models, The Open vs. Closed AI Model Battle, Delaware’s anti Elon ruling, & a Market Update
182 snips
Mar 7, 2024 The conversation kicks off with a dive into the complexities of venture capital and the risk landscape in Delaware. Tesla’s Full Self-Driving V12 is discussed, focusing on its imitation learning techniques and the shift towards neural network models for enhanced driving autonomy. The podcast also highlights the fierce competition between open and closed AI models, emphasizing transparency and regulatory challenges. Finally, recent court rulings affect corporate governance, along with a macro market outlook that scrutinizes major tech companies and their innovation strategies.
AI Snips
Chapters
Transcript
Episode notes
Tesla FSD 12's Imitation Learning
- Tesla's Full Self-Driving (FSD) version 12 uses imitation learning, a radical shift from previous versions.
- The model learns from video in and control signals from expert drivers.
Corner Cases and Deterministic Models
- Previous self-driving car models struggled with corner cases due to complex, deterministic coding.
- Tesla's FSD 12 simplifies this with a neural network model trained on videos of expert driver behavior.
FSD 12's End-to-End Learning
- Unlike older models, FSD 12 doesn't deterministically identify objects like stoplights; it learns driver behavior from pixels.
- This end-to-end learning from raw visual data leads to faster and more human-like driving.