Bloomberg‘s Mark Gurman has shared more details about Apple’s plan to improve its Siri personal assistant by using large language models (LLMs) that will make it more conversational and capable of nuanced reasoning. Apple also continues to work on a ChatGPT rival.
Gurman says Apple is testing a broad range of LLMs of varying complexity, including versions with 3 billion, 7 billion, 33 billion, and 150 billion parameters. As you might expect, the 150 billion parameter model is cloud-based and it is much more powerful than the on-device Aple Intelligence, which uses foundational models with 3 billion parameters.
While Apple has run benchmarks on the model that suggest it “approaches the quality of recent ChatGPT rollouts,” there are still concerns about its tendency to hallucinate. Gurman also says that there are “philosophical differences” among company executives, though he didn’t provide any additional details on what those differences might be.
Gurman also reports that Apple is internally testing a “Knowledge” chatbot model that accesses the internet to gather and use data from multiple sources. The project is being led by Robby Walker, who recently saw Siri removed from his command. Unfortunately, Gurman’s sources tell him the chatbot project is dogged by the same issues that delayed the Siri revamp.
Gurman says other projects are also underway, including:
It isn’t clear exactly when Apple might implement these technologies, and the company is unlikely to announce any launch timelines at WWDC next month, Thanks to the criticism the company received for announcing Apple Intelligence features at last year’s conference that we still haven’t seen launch.