IBM ALTK-Evolve gives customer-service agents learned memory without changing model weights
IBM Research’s approach distills short behavioral rules from a GPT-4.1 ReAct agent’s past attempts and appends them to the system prompt for later conversations.
2 min readRead story



