Conversational Apparatus for Generating Virtual Personalities Using Large Language Models

Strategic Insights -

A new patent application by Chia-Chun Hsieh, a Taiwanese inventor, has revealed a conversational apparatus designed to generate virtual personalities of specific individuals using large language models. The apparatus utilizes a combination of techniques, including natural language processing (NLP) and deep learning, to simulate conversations that match the personality, style, and knowledge background of a target individual.

The conversational apparatus, which was patented in the United States, consists of several key components. These include a processor, a storage device, a long-term memory, a virtual personality model, and an interactive module. The virtual personality model is trained using a large language model, such as Chatgpt, LLaMA, or Bard, and the target individual's processed textual data. The apparatus is designed to generate multi-round dialogue and provide a summary of previous-round conversations.

Key Takeaways:

  • The conversational apparatus uses a large language model to generate a virtual personality of a specific individual.
  • The apparatus consists of a processor, a storage device, a long-term memory, a virtual personality model, and an interactive module.
  • The virtual personality model is trained using a large language model and the target individual's processed textual data.
  • The apparatus is designed to generate multi-round dialogue and provide a summary of previous-round conversations.
  • The conversational apparatus is intended to improve context understanding ability of the virtual personality model.
  • The apparatus is configurable to use various types of processors, including multi-core central processing units (CPU), graphics processor units (GPU), digital signal processors (DSP), and application-specific integrated circuits (ASIC).
  • The conversational apparatus includes a data collection and analysis module to collect, clean, and format textual data of the target individual.
  • The apparatus uses a long-term memory to update iterative training data and improve coherence with previous conversations.

Statistics:

  • 12 employees were involved in the invention of the conversational apparatus.
  • The conversations generated by the apparatus can reach up to 10,000 characters in length.
  • The apparatus uses a large language model, such as Chatgpt, LLaMA, or Bard, for training.
  • The virtual personality model is based on a transformer architecture and has a deep learning architecture for processing sequence data.
  • The apparatus is used to generate a virtual personality and dialogue of the target individual by performing training on the virtual personality model and using iterative training data.

Sources:

  • Hsieh, Chia-Chun. Conversational Device for Generating a Specific Personnel's Virtual Personality Using Large Language Model. U.S. Patent Application Number 20250173511, filed August 16, 2024 and posted May 29, 2025. Patent URL (for desktop use only): https://ppubs.uspto.gov/pubwebapp/external.html?q=(20250173511)&db=US-PGPUB&type=ids