Beyond Games: Can Reinforcement Learning Drive the Quest to Artificial General Intelligence?

From mastering complex video games to defeating the world champion in Go, Reinforcement Learning (RL) has demonstrated its ability to solve problems requiring strategic planning, adaptability, and decision-making under uncertainty. In recent years, RL has expanded its influence beyond games. It has become a critical component in fine-tuning large language models (LLMs), helping them refine their outputs based on user feedback and align with desired behavior. RL is also playing an important role in the development of systems that explore and solve complex coding tasks or mathematical problems, showcasing its versatility across a range of fields. The potential of RL goes even further: as we move towards developing Artificial General Intelligence (AGI), RL's ability to enable continuous learning and adaptability will be essential. It offers a framework for creating systems that can learn autonomously, generalize across tasks, and navigate complex environments—key components on the path to AGI. This talk will delve into the remarkable successes of RL, and its future potential as a cornerstone technology in the quest for AGI. 

We will be serving baguettes during the lecture.

Read more about Alexandre Proutiere.