Apple is reported to be working on its own foundation AI model, codenamed Ajax. This is likely to include the development of a large language model chatbot similar to OpenAI’s ChatGPT, Google’s Bard or Anthropic’s Claude. The company already uses on-device processing to run foundation models for voice cloning and translation.
Engineers inside Apple are said to have been referring to the new model as Apple GPT as it is based on similar technology to the generative pre-trained transformer (GPT) models built by OpenAI. Apple has not commented on the claims, first reported by Bloomberg.
AI, particularly generative AI, is transforming the economy, with companies like Microsoft, Google and Salesforce embedding automated tools across their software portfolios. This week, Meta open-sourced its own Llama 2 large language model, with chipmaker Qualcomm saying the model would be able to run on smartphones from 2024 using the company’s Snapdragon processors.
Apple insiders quoted by Bloomberg said a lot of work is currently being undertaken around addressing potential privacy concerns. Privacy has been a major differentiator with Apple, with Siri, the company’s current AI voice assistant, processing data on-device rather than sending personal information to the cloud.
During its WWDC developer conference earlier this year, Apple unveiled a number of apps and services within its operating system family that rely on AI, but didn’t mention the term itself. Its leaders instead focused on specific phrases like machine learning and foundation models, with a focus on privacy and on-device operation.
The upcoming iOS 17 will include a foundation AI model that runs on the built-in Apple Silicon chip neural engine to transcribe and translate text in real time. This will allow Apple to offer those services without having to send data to the cloud.
CEO Tim Cooke has previously said during an earnings call that AI has potential but there are issues that need to be resolved. It is thought he was talking about privacy concerns. He told ABC’s Good Morning America he uses ChatGPT and the company is “looking at it closely”.
Apple puts AI safety first
The new Ajax framework includes the development of a ChatGPT-like tool that is being used by employees internally. It was said to have been developed last year to allow Apple to unify its machine learning development.
The Bloomberg report suggests Apple built a chatbot on top of the Ajax framework at the end of last year but initially halted development over security concerns. It has since extended the use to more employees to test it for a potential wide-scale rollout. Ajax is said to be built on top of Google Jax, the machine learning framework built by the tech giant. This is likely to be because Apple runs many of its systems on Google Cloud.
However, insiders say there is one caveat over the use of the chatbot – nothing it produces can be used to develop features bound for customers. Employees say its tool works like Bard, ChatGPT and Bing, running as an intranet application, not for public use.
It is also unlikely to be made public or integrated into Siri without significant work on the training data and underlying models. Apple hasn’t revealed details of the size and scale of the underlying models, nor has the notoriously secretive company commented on these reports. Before the public rollout of generative AI, Apple would need to provide the necessary infrastructure and ensure safety requirements are met.
Apple’s head of machine learning, John Giannandrea, and its head of software, Craig Federighi, are said to be heading efforts for the company on how to best deploy and integrate generative AI. Rivals like Microsoft and Google are actively integrating the technology within their productivity suites for business users, including providing generated text from a prompt in Docs and Word, or improving autocomplete within Sheets and Excel.
Read more: Elon Musk launches xAI to take on OpenAI
