Apple’s research team has released their AI innovation that enhances the understanding of contextual queries from users, drawing from the information displayed on users’ screens. This AI system, known as ReALM, short for Reference Resolution As Language Modeling, is an LLM model capable of deciphering unclear or convoluted questions by gathering additional information from the user’s display. This leads to smoother and more natural conversations.
For example, when asking a chatbot to show nearby pharmacies, the usual response would list the names of pharmacies. With ReALM, users can further inquire by saying, “Call the pharmacy on XXX street,” “Call the one at the bottom,” or “Call this number” (displaying a single telephone number). Unlike traditional chatbots that struggle with ambiguous queries, ReALM can continue to work efficiently by utilizing on-screen data. Research comparing this type of questioning with ChatGPT (both GPT-3.5 and GPT-4) found that ReALM performed better.
This research highlights a potential feature that Apple could integrate into Siri in the future.
Source: VentureBeat
TLDR: Apple’s ReALM AI enhances user interactions by understanding contextual queries and utilizing on-screen information, outperforming traditional chatbots in handling ambiguous questions.
Leave a Comment