Conversational Agents

Conversational agents (CA) are engaged in interactive conversations with users, providing responses and assistance while combining Natural Language Processing (NLP), Understanding (NLU), and Generating (NLG) techniques. Two tiers of conversational agent derivation from Large Language Models (LLMs) exist. The first tier involves conversational fine-tuning from datasets, representing expected user questions and desired conversational agent responses. The second tier requires manual prompting by human operators and evaluation of model output, which is then used for further fine-tuning. Fine-tuning with Reinforcement Learning from Human Feedback (RLHF) models perform better but are resource-intensive and specific for each model. Another critical difference in the performance of various CA is their ability to access auxiliary services for task delegation.