[ad_1]
Apple Intelligence has been a wreck since it first deployed last year. Big new report from BloombergMark Garman of , details why. And much of that effort depends on rebuilding Siri from scratch.
Gurman has previously reported that Apple has been working on what it calls “LLM Siri” internally. The previous approach of Apple integrating assistants with existing SIRIs has not worked. Gurman explains in detail several reasons, but here is a brief summary.
The company is currently trying to readjust that approach. Part of that is a complete overhaul of Siri rather than trying to work with the old Siri to make generative AI work. According to Gurman, Apple has a Zurich AI team and is working on a new architecture that “builds completely on an LLM-based engine.” Gurman reported last November that the company was working on this. The idea is to make assistants “in more reliable conversation and information integration.”
Another part of the solution is taking advantage of the differences in privacy to improve synthetic data between iPhone and Apple. Compare fake training data with the language of emails for iPhone users, but use it to send only synthetic data to Apple for AI training. And one way the company is discussing improvements to Siri is to loosen the LLM version on the web and “grab and synthesize data from multiple sources.” Essentially, AIL is Siri as an AI web search tool, unlike Prplexity, one of the companies Apple approached for AI search partnerships in Safari.
Whatever the outcome, Giannandrea appears to be not a direct part of it, being excluded from product development, Siri and robotics projects in the spring. According to Gurman, Apple executives have spoken about putting him on the “road to retirement,” but are concerned that some of the research and engineering people he brought with him will also leave with him. In any case, Gurman says Giannandrea will stick with, “Reassuring Siri is now a problem for others.”