Apple GPT will be on the iPhone next year! Apple was exposed to secretly developing the Ajax framework, Siri was greatly upgraded, and the market value soared by hundreds of billions in seconds

Original source: Xinzhiyuan

Image credit: Generated by Unbounded AI‌

Apple, which is "staying out of the way", is finally going to scale up its moves.

According to Bloomberg, Apple is secretly developing and testing artificial intelligence tools, and insiders call it "Apple GPT", directly competing with OpenAI and Google.

The development framework of "Apple GPT" is called Ajax because it is built using the Google JAX framework.

Apple's goal is to launch consumer-grade products in 2024.

In addition, what is certain is that the mentally handicapped Siri should be saved!

As soon as the news came out, Apple's market value surged by nearly 70 billion U.S. dollars (over 500 billion yuan) in a few seconds, and its stock price continued to hit a record high.

In the AI large-scale model arena, OpenAI took the lead with GPT-4 momentum, followed by Google PaLM, and Meta opened up an open source exclusive route.

Microsoft has closed source GPT-4 in one hand and open source Llama 2 in the other, making a lot of money.

This time, the addition of Apple means that the artificial intelligence large-scale language model battlefield is about to set off a frenzy of great changes.

As netizens said, there are 1.5 billion iPhone active users there, and Apple can change the LLM pattern overnight.

** "Apple GPT" Secret Research and Development**

Back in the last year, Apple created the Ajax framework for the first time.

The system is built on top of Google Jax and aims to unify Apple's machine learning development.

Ajax is now being used to create large language models and as the basis for an internal ChatGPT-like tool "Apple GPT".

It is said that Apple has made artificial intelligence improvements to search, Siri, and maps based on this system.

The biggest use case for Apple's LLM technology would be to integrate it within Siri, allowing the voice assistant to perform more tasks on behalf of the user, the people said.

The effort is led by John Giannandrea, head of machine learning and artificial intelligence, and Craig Federighi, head of software engineering.

But they have yet to form a united front within Apple.

Giannandrea said he wants to take a more conservative approach, hoping to see how recent developments at other companies have evolved.

In addition, around the same time that Apple began developing its own tools, Apple conducted an enterprise trial of OpenAI's technology.

It is also considering a larger contract with OpenAI. Previously, OpenAI has licensed its services to Microsoft, Shutterstock and Salesforce.

Not only that, but Apple is hiring more generative AI experts.

On its website, it's hiring engineers with "a deep understanding of large-scale language models and generative AI."

Apple has been incorporating AI elements into each of its products for years. But in the current explosion of generative AI, Apple is catching up.

In the past year, after ChatGPT was born, AI tools such as Google Bard and Microsoft Bing were launched one after another, and generative AI has become mainstream.

However, in the eyes of the outside world, Apple did not declare war on a large scale to participate in this fanatical model war.

It is undeniable that Siri, Apple's main artificial intelligence product, has stagnated for many years and has been criticized.

I still remember that in 2011, on the day when the iPhone 4 came out, Siri made its official debut as an intelligent voice assistant, and everyone on the spot was amazed by it.

Beyond Siri, though, Apple has used artificial intelligence to improve other areas of its products, such as photos and iPhone search.

In addition, iOS 17, which will be launched in the second half of this year, can also achieve smarter automatic correction.

People familiar with the matter said that in recent months, Apple has focused on advancing artificial intelligence research and convened several teams to work together to tackle key problems.

The work is said to include an attempt to address potential privacy issues associated with “Apple’s GPT” technology.

Earlier, Cook said in an interview that he used ChatGPT, a chatbot, and was excited about it. He also revealed that Apple is paying close attention to this technology and has some unique application scenarios.

Cook has repeatedly stated in public that Apple recognizes the potential of artificial intelligence, but there are still many problems to be solved, and how to use the technology requires "deliberate consideration."

Indeed, inside Apple, executives are increasingly worried about missing out on a major shift in how products work.

Because generative AI promises to change the way people interact with phones, computers and other technologies.

If Apple doesn't keep up with advances in artificial intelligence, product sales could suffer. Apple's products generated nearly $320 billion in revenue last fiscal year.

But the question now is how to commercialize the AI model?

**With AI weapons, how to use them? **

At the end of last year, a small number of Apple engineers experimentally created "Apple GPT".

However, due to concerns about the security of generative AI, "Apple GPT" was blocked at the beginning of Apple's launch, and it was only tested among some internal employees.

Some employees even had access to the system only with special approval.

And, Apple seriously warns: Any output from "Apple GPT" cannot be used to develop customer-facing features.

Even so, Apple employees still use it to assist with product prototyping. "Apple GPT" also summarizes text and answers questions based on the training data.

In fact, Apple's caution is not the only one. Other technology companies, including Samsung, have also developed their own ChatGPT-like tools to prevent third-party services from leaking sensitive data.

Apple employees said that the function of "Apple GPT" is basically a "replica" of Bard, ChatGPT and Bing AI, and does not include any new features or technologies.

The system is accessible as a web app, but its stripped-down design is closed to the public.

As such, Apple currently has no plans to release it to consumers, although it is actively working to improve its base model.

Beyond the current state of the technology, Apple is still trying to determine how generative AI should be aimed at consumers.

It is now working on several related initiatives:

A cross-company collaboration between the AI and software engineering teams, and the cloud services engineering team that will provide the infrastructure for any significant new AI capabilities.

Apple is aiming for a major AI-related announcement in 2024, people familiar with the matter said.

Cook defended the company's artificial intelligence strategy in a May conference call, saying the technology is used in most of its product lineup, including features such as car crash and fall detection.

Cook defended the company's artificial intelligence strategy in a May speech, saying the technology is used in much of its product lineup, including features such as car crash and fall detection.

More recently, Cook said the LLM is likely to warn about both bias and disinformation. He also called for guardrails and regulation in the area.

Siri is too stupid, all because of the "shit mountain" is difficult to deal with

This revelation further confirms how important Apple's internal bet on generative AI is.

In fact, foreign media also reported that Apple has begun to test "language generation AI" internally, all because Siri is too far behind.

Apple engineer John Burkey commented in an interview with The New York Times:

Siri is too dumb to compete with ChatGPT! It can't be a "creative assistant" like ChatGPT, and the clumsy code makes it difficult to add new features.

Burkey worked on improving Siri in 2014 and left Apple in 2016.

Siri, what's the weather like?

Siri, can you sing this song?

Such a simple question, Siri can do it, mainly because it is done by extracting a large number of words from the database.

This means that Siri can only understand some human requests. Engineers had to add new words to the database in order to expand its functionality.

But Siri was essentially a "command-and-control" system, and its clumsy design made it difficult for engineers to add new features.

Siri's database contains extensive lists of phrases in nearly 24 languages, making it "one giant snowball," or what is often referred to as the mountain of ancestral code shit.

It may take about a year to integrate more advanced features like ChatGPT, such as new search tools.

What's more, even upgrading Siri's basic functions could take weeks due to the intricacies of the code.

In February, Apple held its annual AI summit, focusing on current AI tools and large language models.

Many teams, including engineers working on Siri, are regularly testing "language generation concepts" "on a weekly basis," the New York Times said.

In addition, Apple has tested a new framework for "Siri Natural Language Generation" in tvOS 16.4, internally codenamed "Bobcat".

In addition to testing language generation AI, in January this year, Apple launched a plan to provide digital narration on Apple Books, which can automatically generate high-quality artificial intelligence narration audio from written text.

This suggests that Apple is already exploring use cases for generative AI.

This model war is waiting for Apple to play.

References: