Apple has built its own large language model (LLM) AI framework and associated chatbot service similar to OpenAI’s ChatGPT, Google’s Bard, and Microsoft’s ChatGPT-based Bing Chat, according to a new report.
The project is codenamed “Ajax” internally, the sources say, though some engineers within the company colloquially refer to it as “Apple GPT.” It is built on top of Google machine learning framework Jax and runs on Google Cloud.
The news comes via an article from Bloomberg’s Mark Gurman, who cites “people with knowledge of the efforts.” Gurman has generally been accurate in his reports citing sources like these in the past, though he has seemingly misinterpreted what he was told on rare occasions.
Ajax was developed “as an experiment” around the end of 2022 and has since been used by Apple employees for some tasks like product prototyping. However, there have been stops and starts. Initially, the rollout was put on hold “over security concerns,” and there is a rule within the company that it can’t yet be used to build anything that will ultimately be consumer-facing.
Employees must go through an application and approval process to begin using it. Once they gain access, they can use it as a web app, albeit one that has “a stripped-down design not meant for public consumption.”
All that said, Apple hasn’t finalized a plan for rolling out features driven by Ajax to consumers. Apple execs John Giannandrea and Craig Federighi are leading teams that are working on figuring all that out—reportedly with the intent to announce something in 2024.
Giannandrea has been leading Apple’s AI efforts since he defected from Google in 2018, but Apple’s AI strategy has focused on conservative applications that use on-device processing to improve the user experience or introduce extra features. That lies in stark contrast to the potentially more disruptive experimentation at companies like Google or OpenAI.
A report earlier this year from The Information indicated that Apple was nonetheless working on something using LLMs, though it wasn’t very specific about it. It indicated that Apple leadership was hesitant to go all-in on the technology due to fear of the unforeseen consequences of LLM-based chatbots and features giving incorrect information or “hallucinations,” as AI researchers call those sorts of errors. And it also came to light recently that Apple restricted ChatGPT use among employees.
Additionally, most LLMs run in the cloud (as does Ajax), but for various reasons, Apple has preferred to keep most AI-based features running on users’ local devices.
The Information’s report also claimed that Apple was facing a brain drain of strong AI talent to companies like Google and Microsoft with less conservative attitudes about AI.
All that is to say that even though Apple is working on and playing with ChatGPT-like AI tools within the company, there are some major barriers to bringing those tools to Apple products, and it’s still unclear how (or even if) Apple will try to break through those obstacles.
https://arstechnica.com/?p=1955544