Apple’s (NASDAQ:AAPL) latest research paper suggests the tech giant will be an “AI winner” when it solves the issue of having on-device inference that can enable an iPhone “supercycle,” Evercore ISI analysts said.
The research, which was published on Monday, describes Apple’s newest language model, known as Reference Resolution As Language Modeling, or ReALM. The researchers indicated the smaller ReALM models were similar in performance to OpenAI’s GPT-4, albeit with fewer parameters. This could allow the new model to deliver AI to the edge, or devices such as smartphones and PCs and provide the same function as AI in data centers.
“The key innovation here appears to be around reference resolution, which refers to how AI can understand references to on-screen data as well as more conversational context,” Evercore ISI analyst Amit Daryanani posited. “Apple simplifies the normally complex task of reference resolution by converting all on screen data to text, which is far easier for an [large language model] to parse relative to images. This is the key feature that enables Apple to deliver high performance with a more limited number of parameters.”
Daryanani maintained his Outperform rating and $220 price target on Apple.
By offering AI tools on consumer devices (and perhaps increasing the usefulness of Siri), Apple could offer developers lower latency, better security and make it easier and cheaper to do AI-related work, Daryanani said.
“We think AAPL’s AI strategy will focus on incorporating on-device inference for [large language models] that will substantially uplift the user experience for not only the iPhone but also Mac/iPad,” Daryanani explained. “Given their vertical integration and especially their control over their own silicon, AAPL is best positioned to not only expand the moat surrounding the iOS ecosystem but also potentially drive an accelerated refresh cycle should the final implementation be deemed a big enough change.”
He continued: “The refresh cycle will likely be driven by significant updates to the hardware including a potentially improved neural engine or GPU. Additionally, a significant enough upgrade in capabilities could potentially drive an uplift in [average selling prices].”
Apple is slated to kick off its annual developer conference on June 10 where many are expecting the tech giant to make a number of AI-related announcements.
Source link
credite