D ShinFollow


Virtual assistants, chatbots, and other natural language conversational agents enable users to provide spoken or text queries in a natural manner and receive responses. Such conversational agents can utilize large language models (LLMs) as a backend mechanism for query interpretation and response generation. When a user pauses or otherwise interrupts the query input, or modifies query input, the LLM may generate and provide a response to the original query rather than the user’s intended query. This disclosure describes techniques that enable on the fly query interpretation and updates to response generation when a user in conversation with a conversational agent appends or modifies a query, even as response generation is underway. If further user input is detected, the response generation operation is aborted. The LLM prompt is updated to include the original query with the updated input, e.g. by concatenation or modification, and the updated response generated by the LLM is provided in the conversational interface.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.