Computer Use Agents: From SFT to Classic RL [ukr]
Let’s explore Computer/Browser/Mobile Use agents. We’ll start with the APIs provided by OpenAI and Claude that support such use cases. Then, we’ll recall how LLMs and VLMs are trained, what Reinforcement Learning (RL) is, and how it can be applied in this context. We'll also look into some recent open-source agent models, and discuss how to evaluate these agents effectively.
Maksym Shamrai
Research Scientist at MacPaw
- Research Scientist at MacPaw AI Research (AIR), working on applied research in the field of AI
- PhD student at the Institute of Mathematics of the National Academy of Sciences of Ukraine, researching optimization methods for neural network models and optimal control theory
- Teaches "Computer Vision" to Master's students at Kyiv Academic University (KAU)
- Interested in the mathematics behind AI, particularly in Reinforcement Learning
- Has experience in both research and production