An Empathic Large Language Model (eLLM), as defined by Hume AI, is a cutting-edge AI technology that aims to significantly enhance the accuracy of predictions related to human preferences, needs, and overall well-being by incorporating multimodal signals. This integration includes language, voice, and facial movements into a custom model, leveraging Hume's AI models pre-trained on a vast collection of videos and audio files. The result is a system that can predict outcomes with high accuracy, often after being exposed to just a few dozen examples.

Hume's approach with eLLM is distinctive because it combines these diverse signals with language, embedding them into the eLLM through pretraining on millions of human interactions. This method allows for the creation of custom models using Hume's Custom Model API that are uniquely adept at understanding and predicting nuanced human expressions and emotions. An example of its application includes a partnership with Lawyer.com, where using just 73 calls, Hume was able to train a model to predict the quality of customer support calls with 97.3% accuracy, a significant improvement over models using language alone, which had a 3x higher error rate.

This advancement in empathic AI suggests a paradigm shift in how technology can interpret and respond to human emotions, offering applications across a wide range of fields from customer service to mental health assessment.

Try Hume's conversational Empathic Voice Interface (EVI) here: https://demo.hume.ai

Want to explore the foundations of large language models like ELLM? Enroll in the Generative AI: Foundation Models and Platforms course on Coursera to dive deep into the world of generative AI and platform technologies.