AI-modellen trainen met productiedata: van OpenTelemetry naar slimmere taalservers

Researchers are developing smaller language models trained on production data from AI agents, using OpenTelemetry to capture user interactions as training signals. This approach enables organizations to create cost-effective custom models that replicate the behavior of larger frontier models without requiring expensive infrastructure.