Apple is making strides in generative artificial intelligence by conducting research to run large language models (LLMs) on smartphones, a departure from the traditional cloud-based approach. A paper titled “LLM in a Flash,” published on December 12, demonstrates a solution for running LLMs effectively on devices with limited memory, such as iPhones. This development addresses the computational challenges of operating sophisticated AI models like ChatGPT on mobile devices.
The paper, which gained attention after being featured by Hugging Face, signifies Apple’s effort to catch up with Silicon Valley peers in AI innovation. Apple’s previous initiatives, like the integration of image-generating models on custom chips, already show its commitment to AI advancement.
This research is timely, as the smartphone market experiences a downturn, with a 5 percent decline in shipments, according to Counterpoint Research. Apple, despite pioneering with Siri, has been perceived as lagging in the generative AI race, a gap it’s now addressing by exploring AI functionalities independent of cloud computing.
Unlike competitors focusing on cloud-based AI services, Apple is steering towards device-based AI. This aligns with its emphasis on user privacy, as AI processing on personal devices potentially enhances data security. Apple’s approach involves optimizing smaller models, like Falcon 7B, for efficient inference on battery-powered devices.
This shift in AI strategy could redefine the smartphone experience, offering quicker, offline-capable AI assistance and advanced privacy measures. The paper’s insights, while not direct indicators of Apple’s product plans, reveal a significant move towards leveraging AI’s potential in various devices and applications, setting a precedent in the industry. Apple’s initiative underscores the growing trend towards integrating sophisticated AI features directly into consumer devices, a move expected to shape future technological advancements and user experiences.
