Zhitong
2024.04.02 23:40
I'm PortAI, I can summarize articles.

Winning without running? Apple releases latest large-scale paper, Evercore says it is expected to become the "AI winner"

Apple's latest research report indicates that it will emerge as the winner in artificial intelligence once the inference problem on devices is solved. The study describes Apple's latest language model, ReALM, which performs similarly to OpenAI's GPT-4 but with fewer parameters. This allows the new model to provide artificial intelligence on edge devices and have the same capabilities as AI in data centers. Analysts believe that Apple has simplified the reference parsing task, enabling high-performance key features with limited parameters

According to the Zhitong Finance and Economics APP, Evercore ISI analysts stated that Apple (AAPL.US) latest research report indicates that when this tech giant solves the reasoning problem on devices that can achieve an "iPhone super cycle", it will become the "winner of artificial intelligence".

This research was published on Monday, describing Apple's latest language model, ReALM (Reference Resolution As Language Modeling). Researchers stated that the smaller ReALM model performs similarly to OpenAI's GPT-4 in terms of performance, despite having fewer parameters. This allows the new model to bring artificial intelligence to edge devices such as smartphones and personal computers, providing the same functionality as artificial intelligence in data centers.

"Our model has made significant progress in identifying different types of entities, with even the smallest model achieving an accuracy improvement of over 5% in screen entity recognition compared to the original system. In comparison with GPT-3.5 and GPT-4.0, our smallest model performs similarly to GPT-4.0, while larger models outperform it," the researchers stated.

Evercore ISI analyst Amit Daryanani believes, "The key innovation here seems to revolve around reference resolution, that is, how artificial intelligence understands references to screen data and more conversational context. By converting all data on the screen into text, Apple has simplified the typically complex reference resolution task, which is easier to parse for (large language models) compared to images. This is a key feature that enables Apple to provide high performance with limited parameters."

Daryanani maintains his "outperform" rating on Apple and a target price of $220.

Daryanani stated that by providing artificial intelligence tools on consumer devices (perhaps also improving the utility of Siri), Apple can offer developers lower latency, better security, and make AI-related work easier and cheaper.

Daryanani explained, "We believe that Apple's artificial intelligence strategy will focus on integrating reasoning on devices with large language models, which will greatly enhance the user experience of iPhone and Mac/iPad. Considering their vertical integration, especially their control over their own chips, Apple is in the best position to not only expand the moat around the iOS ecosystem, but also potentially accelerate the upgrade cycle if the final implementation is deemed a significant enough change." The analyst added: "The update cycle may be driven by significant hardware updates, including potentially improved neural engines or GPUs. In addition, if the features receive a significant upgrade, it may drive up the average selling price."

Apple is scheduled to hold its annual developer conference on June 10th, and many expect the tech giant to make a series of announcements related to artificial intelligence