AI Learning & Development

Agent experiments and results can be stored for future reference and used in two distinct ways:

1. Past experiments and a record of their success or failure are stored to a RAG database. When the agent encounters a similar situation in the future, the relevant information is retrieved and used to inform its decision-making.

2. Past experiments and a record of their success or failure are saved to a DPO or a GRPO dataset to retrain the model.

While our initial experiments in this field have focused on transformer models for ease of use, in future our aim is to use text diffusion models for improved generalisation.