Dropping from 54GB to 4GB, Apple is in talks with PrismML. Is model compression technology about to boom? What? Large language models are finally getting "weight loss shots" too? This isn't just made up. According to CNBC reports, Apple is in talks with the startup PrismML, a company renowned for its recently launched model compression technology. Apple aims to evaluate the feasibility of running larger-scale AI models directly on iPhones using this technology. (Image Source: CNBC) You know, over the years, whenever the AI segment comes up at smartphone launch events, I usually instinctively reach for my water bottle. It's not that I have any grudge against manufacturers. It's just that everyone is far too familiar with this routine. First, let the AI summarize various elements on the screen, then use image editing tools for personalized color grading or erase passersby from photos. This year, a new feature has been widely added: calling a voice assistant to order you a coffee. But we can't blame smartphone manufacturers entirely. Current mainstream large models simply can't fit into mobile phones, and the trimmed-down on-device models lack sufficient intelligence. In the end, all manufacturers can promote are cloud-updated features. For example, once Doubao launched an AI podcast function, almost all mainstream manufacturers followed suit within three months. Here's the question: if a full large model is slimmed down enough to fit on a phone, can on-device AI assistants finally become fully functional? From 54GB to 4GB: Is model compression technology about to go mainstream? First, let's explore these two questions together with Leitech (ID: leitech): Who is PrismML? According to its official website, PrismML is a startup specializing in model compression. It spun off from a California Institute of Technology research team and is backed by Khosla Ventures, Cerberus, and Google. Its research