Abstract
A low-intelligence, or small, large language model (LLM), termed a Vibe LLM, operates on resource-constrained hardware to augment user experiences. The Vibe LLM does not handle direct linguistic interaction with a user; instead, it processes textual output generated by a separate, more resource-intensive LLM that typically resides in the cloud. By analyzing this text, the Vibe LLM extracts contextual information, such as proper nouns, and uses it to control device features like light emitting diode (LED) animations or display colors. For example, the Vibe LLM can identify a team name mentioned in a response and trigger an LED animation using that team’s colors while the audio response plays. A larger LLM generates a diverse training dataset of lookup tables and sample responses, which a training process then uses to fine-tune and compress the Vibe LLM for on-device deployment, enabling dynamic, context-aware device behavior without needing the hardware to run a large-scale conversational model locally.
Keywords: large language model (LLM), on-device LLM, Vibe LLM, user experience augmentation, LED animation, home device settings, text-to-speech (TTS), smart speaker, color customization
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Mirsky, Yosef and Felch, Andrew, "On-Device Large Language Model for Enhanced User Experience Customization", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/10416