A group of tech workers in San Francisco, calling themselves DrivingBench, successfully demonstrated that general-purpose large language models can be trained to drive a car. They modified a Toyota Corolla using an open-source system called Comma, allowing LLMs like GPT-6 Astra to control the vehicle's steering, acceleration, and braking. While some models struggled, GPT-6 Astra was able to navigate a cone course in a parking lot, proving the potential for LLMs to perform real-world robotic tasks. AI
IMPACT Demonstrates the potential for general-purpose LLMs to perform complex real-world tasks, potentially accelerating the development of AI-powered robotics and autonomous systems.
RANK_REASON Demonstration of LLM capabilities in a real-world robotic task using off-the-shelf hardware and software.
- Aditya Ramabadran
- Axiom Math
- ChatGPT
- Claude Fable 5-1
- Comma
- DrivingBench
- GPT 5.6 "Sol"
- GPT-6 Astra
- Grok 4.6
- San Francisco
- Simon Mahns
- Tesla
- Tobias Gessler
- Toyota Corolla
- Waymo
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
