Vai al contenuto
AI.info

events

Closing the Continuous Learning Loop: From Agent Traces to Agent Training

Rustem Feyzkhanov spiega come le tracce degli agenti LLM in produzione possano alimentare simulazioni riproducibili, ricompense verificabili e fine-tuning o apprendimento per rinforzo.

Closing the Continuous Learning Loop: From Agent Traces to Agent Training | Eventi sull’IA

In inglese

Starts
2026-10-19 16:00 UTC
Organizer
AI Performance Engineering Meetup (San Francisco, Global)
Location
Online event

An online session from AI Performance Engineering Meetup, held over Zoom. After brief introductions and group updates from Chris Fregly and Antje Barth, Rustem Feyzkhanov of Snorkel.ai discusses a practical loop for using production agent traces: turning failures into reproducible simulations and verifiable rewards, then improving models through fine-tuning or reinforcement learning. The talk also compares when prompt, tool or agent-harness changes are preferable to updating model weights. It is aimed at engineers and researchers working on LLM agents and model improvement.

Event website