Apple’s new artificial intelligence: 3 ways Siri beats ChatGPT

Apple seems slow to jump on the generative AI bandwagon, but it’s new Research Something about contextual understanding might make Siri better than ChatGPT.

The tech giant has been unusually quiet during a period of rapid growth The rise of ChatGPT and a subsequent set of generative AI tools and capabilities from the company Google, Microsoftand YuanBut Apple researchers have a new model that could give Siri the generative artificial intelligence upgrade Apple fans have been hoping for.

“Human speech often contains vague references, such as ‘they’ or ‘that’, whose meaning is obvious (to others) given the context,” the researchers said. The paper proposes a The model, called ReALM (Reference Resolution as Language Modeling), solves the problem that large language models (LLMs) do not always work when it comes to on-screen, conversational, and background references (e.g., apps or features running in the background). It’s a matter of being able to understand the context, with the goal of achieving “real hands” “-experience the voice assistant for free.”

While ChatGPT is very good and can understand certain types of context, the researchers said ReALM outperformed GPT-3.5 and GPT-4 (supported by both free and paid versions of ChatGPT) in all context tests. What this could mean for Siri.

1. On-screen context clues

Apple researchers trained ReALM using “on-screen” data from web pages, including contact information, allowing the model to understand text in screenshots (such as addresses and bank account details). While GPT-4 can also understand images, it can’t. The paper believes that this allows ReALM to better understand the screen information that Apple users will ask Siri for help.

2. Dialogue and contextual understanding

A dialogue reference means something that is relevant to the dialogue but may not be explicitly mentioned in the prompt. By training ReALM using data such as business lists, the model can understand prompts, such as “Call the one at the bottom,” referring to the list of nearby pharmacies displayed on the screen, without providing more specific instructions.

ReALM understands “background entities,” meaning that something running in the background of a device “may not necessarily be a direct part of what the user sees on the screen or interacts with the virtual agent,” such as music playing or an alarm going off.

3. Completely on the device

Last but not least, ReALM is designed to run on-device, which will be a big deal since LLM requires a lot of computing power and is therefore primarily cloud-based. In contrast, ReALM is a smaller LLM, but fine-tuned for specific cases and explicitly targeted at the task of reference resolution. Apple has historically made its commitment to privacy a selling point of its devices, so a generative AI version of Siri running entirely on-device would be very on-brand and a major achievement for an AI-capable device.

As expected, Apple has been tight-lipped about its artificial intelligence plans, but CEO Cook said Expect big announcements on artificial intelligence All eyes will be on Apple’s Worldwide Developers Conference (WWDC) later this year June 10.


Source link

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button