Stephanie Zhan What do you think were the breakthroughs that were made at that time? Also, what do you think were the main challenges facing the field of intelligent agents at that time? Jim Fan Yes, the main method that we used at that time was reinforcement learning. There was no LLM or Transformer model in 6th. Reinforcement learning works on specific tasks, but it does not generalize broadly. For example, we cannot give any instructions to the agent and ask it to perform different tasks that are controlled by the keyboard and mouse.
At that time, it worked on the specific tasks that we designed cameroon phone numbers it for, but it did not generalize at all. University. I started my PhD with Professor Favilli at Stanford, focusing on computer vision and embodied intelligence (Embodied AI). During my time at Stanford from 6th to . In 2011, I witnessed the transformation of the Stanford Vision Lab from static computer vision led by Professor Favilli, such as image and video recognition, to embodied computer vision, that is, agents in interactive environments that learn to make sense and act.
This environment can be virtual (in a simulation) or in the physical world. So this is my PhD phase, which is mainly about moving from static vision to embodied intelligence. After my PhD, I joined Nvidia and have been working there ever since. I brought the research content from my PhD thesis to Nvidia and continue to work on embodied intelligence to this day. Sonya Huang You are currently responsible for Nvidia's Embodied Intelligence initiative, can you briefly describe what this project is and what you hope to achieve? Jim Fan Absolutely.
That led me to the next level, which was to go to Stanford
-
- Posts: 703
- Joined: Fri Dec 27, 2024 12:33 pm