Learning Loop

Definition. The mechanism by which an agent improves from use: human edits are diffed against the original, structured observations are extracted and stored (e.g. per-rep style memory), and every future run reads them before acting. Human-in-the-loop doubles as the data-collection engine — the loop that turns the execution tier into a learning system.

@misc{shalvi_gtm_learning_loop,
  author = {Singh, Shalvi},
  title  = {Learning Loop — GTM World Model Glossary},
  year   = {2026},
  url     = {https://shalvisingh.com/gtm/glossary/learning-loop}
}