Taxonomy of Self-evolving Agents
What does it mean for an AI agent to improve itself—and what is actually changing?
Join the Nebius Science Paper Club for a session with Shilong Liu (Princeton AI Lab) exploring three directions in self-evolving agents: improving the outputs they produce, updating their prompts, memory, tools and skills, and training the underlying model without gold-standard answers.
The talk will introduce a framework for comparing these approaches, understanding where they overlap, and connecting them to continual learning, test-time training and recursive self-improvement. Bring your questions for Q&A and open discussion.
Read the related blog post before the webinar →
What we’ll cover
- Improving agent outputs: Iteratively refining code, algorithms, scientific findings and robot policies.
- Improving the agent harness: Updating prompts, memory, tools and reusable skills without changing model weights.
- Learning without gold-standard answers: Using self-training, self-play and weak feedback from the environment.
- Connecting research directions: Where these approaches overlap and how they relate to continual learning, test-time training and recursive self-improvement.
Who should attend
ML researchers, research scientists, ML engineers and AI developers building or evaluating LLM-based agents, as well as technical founders and AI team leads exploring agent development and training.
About Nebius Science Paper Club
Nebius Science Paper Club is a webinar series led by Nebius researchers. Each session brings together paper authors and practitioners to discuss new ideas, research, and discoveries in AI.
The session is open to researchers, engineers, students, and anyone interested in knowledge distillation, model compression, and computer vision.
Shilong Liu
Register to get the link and the recording
Try Nebius AI Cloud console today
