share_log

未跑先赢?苹果发表最新大模型论文,Evercore称其有望成为“AI赢家”

Win before you run? Apple published its latest big model paper, Evercore says it is expected to become an “AI winner”

Zhitong Finance ·  Apr 3 08:29

Source: Zhitong Finance

Evercore ISI analysts say Apple's latest research report shows that when the tech giant solves inference problems on devices that can enable the iPhone's “supercycle,” it will be a “winner in artificial intelligence.”

Evercore ISI analysts said,$Apple (AAPL.US)$The latest research report suggests the tech giant will be the “winner of artificial intelligence” when it solves the problem of reasoning on devices that can enable the iPhone's “supercycle.”

The study was published on Monday and describes Apple's latest language model, RealM (Reference Resolution As Language Modeling, that is, reference analysis based on language modeling. Researchers say the smaller Realm model is similar in performance to OpenAI's GPT-4, albeit with fewer parameters. This allows new models to bring artificial intelligence to devices such as edge or smartphones and personal computers, and provide the same functionality as artificial intelligence in data centers.

“Our models have made significant progress in recognizing different types of entities. Even with the smallest model, the accuracy rate of screen entity recognition has improved by more than 5% compared to the original system. “Compared to GPT-3.5 and GPT-4.0, our smallest model is comparable in performance to GPT-4.0, while the larger model is clearly superior to it,” the researchers said.

Evercore ISI analyst Amit Daryanani said, “The key innovation here seems to revolve around reference analysis, that is, how artificial intelligence understands references to screen data and more context of conversation. By converting all the data on the screen into text, Apple has simplified the usually complicated task of reference parsing, which (for large language models) is easier to parse than images. This is a key feature that enables Apple to deliver high performance with a limited number of parameters.”

Daryanani maintained his “outperforming the market” rating for Apple and his target price of $220.

Daryanani said that by providing artificial intelligence tools on consumer devices (which may also improve the utility of Siri), Apple can provide developers with lower latency, better security, and make AI-related jobs easier and cheaper.

Daryanani explained, “We believe Apple's artificial intelligence strategy will focus on integrating inference on devices (large language models), which will greatly enhance the iPhone and Mac/iPad user experience. Given their vertical integration, and particularly their control over their own chips, Apple is in the best position to not only expand the moat around the iOS ecosystem, but if the final implementation is seen as a big enough change, it could also speed up the update cycle.”

The analyst added, “The update cycle is likely to be driven by major hardware updates, including possible improvements to neural engines or GPUs. Also, if the features are upgraded big enough, it may drive up (average selling price).”

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

Editor/jayden

The translation is provided by third-party software.


The above content is for informational or educational purposes only and does not constitute any investment advice related to Futu. Although we strive to ensure the truthfulness, accuracy, and originality of all such content, we cannot guarantee it.
    Write a comment