Apple is developing its own AI-powered chatbot, called “Apple GPT,” which uses its own large language model (LLM) framework called “Ajax” on Google Cloud. The chatbot is built with Google JAX, a framework designed to accelerate machine learning research. Apple has multiple teams working on the project, including addressing potential privacy implications.
In 2018, Apple hired John Giannandrea, who previously headed AI and search at Google, to oversee Siri and its machine learning teams. Giannandrea and Apple’s senior vice president of software engineering, Craig Federighi, are leading Apple’s AI initiative. Apple is closely monitoring AI tech and has expressed concerns about AI products during an earnings call in May. Other companies have begun working together to release different LLMs to startups and researchers, with Meta releasing its LLM LLaMA 2 on Microsoft’s Azure platform. While Apple’s plans in the AI space are still unclear, sources tell Bloomberg that Apple is looking to make a “significant AI-related announcement” sometime next year.
Reports have emerged that Apple is working on a language model, tentatively named “Apple GPT,” possibly as a counterpart to ChatGPT. This news influenced Apple’s market valuation, seeing an increase of $71 billion, highlighting the current interest in AI developments.
Several insights about Apple GPT include:
- Prototype Availability: Apple has internally developed a chat prototype based on the “Apple GPT” which is now available for its developers. Concurrently, in May, Apple directed its employees to abstain from using ChatGPT, emphasizing their commitment to the new initiative.
- Training Infrastructure: To facilitate the training of expansive language models, Apple established their Ajax framework, utilizing Google’s JAX as its foundation.
- Release Timeline: Anticipations suggest that the commercial debut of this technology won’t occur before 2024.
- Terminology Shift: Apple exhibited its characteristic discretion when references to “AI” were replaced with “ML” during the launch of their AR helmet at WWDC.
- Tim Cook’s Remarks: Apple CEO, Tim Cook, in a recent investor call, indicated Apple’s cautious approach toward AI integration, highlighting unresolved issues likely pertaining to data privacy and model inaccuracies.
Apple has restricted employees from using AI tools like OpenAI’s ChatGPT and GitHub’s AI programming assistant Copilot due to concerns about leaked or collected confidential information. OpenAI stores all interactions between users and ChatGPT, which are collected to train its systems and can be inspected by moderators for violating the company’s terms and services. Despite enabling a feature to turn off chat history, OpenAI still retains conversations for 30 days with the option to review them for abuse before permanently deleting them. Apple may be concerned about employees entering confidential project information into the system, which could be seen by OpenAI’s moderators. Research shows it’s possible to extract training data from language models using its chat interface, but there’s no evidence that ChatGPT itself is vulnerable to such attacks.
Apple’s ban is notable given OpenAI’s recent launch of an iOS app for ChatGPT, which is free to use, supports voice input, and is available in the US. OpenAI plans to launch the app in other countries soon, along with an Android version.
While the exact applications of Apple GPT remain undisclosed, there’s speculation about its incorporation into Siri, given the lack of major advancements in the voice assistant in recent times. Given Apple’s strength in hardware and Edge devices, coupled with a less dominant presence in large-scale cloud computing, it remains to be seen if Apple plans on running large language models directly on mobile devices.
Source: mPost
