Nevertheless, Apple was captivated by the potential, as was the then-CEO Steve Jobs. “This was Steve’s last deal,” remarked Tom Gruber, cofounder of Siri, in an interview with WIRED. “He was directly involved in every stage of the negotiation, ensuring we thrived at Apple post-acquisition.”
However, various Apple executives from that period illustrate a picture of a notably imperfect digital assistant that often failed to deliver on what Apple had promised. The initial version of Siri functioned, yet only within highly constrained parameters.
“What we bought was a showcase that would impress a few individuals but wouldn’t adapt to our extensive user base … there was a lot of illusion surrounding the original Siri setup,” shared former Apple executive Richard Williamson during a lengthy 2017 interview with the Computer History Museum, which even included costume changes.
“This concept of AI? It wasn’t genuine AI … it was chaotic,” Williamson stated. “Siri is easily misled. It lacks NLP [natural language processing], and there’s no contextual understanding of language. It’s merely matching keywords.”
Yet, even with advancements in AI, Siri reportedly remains unreliable when confronted with real-world usage. The pressing question becomes: why? Although chatbot technology may not be fully evolved, it is common enough to be utilized daily by everyday users on rival platforms.
One perplexing element: Apple’s methodology in this area likely diverges from industry norms. Using Alexa effectively requires users to be comfortable sharing extensive data, while OpenAI’s Sam Altman appears willing to overhaul entire job sectors in pursuit of progress. In contrast, Tim Cook and Apple? A cleaner, more altruistic image has been integral to the company’s allure for decades, which includes a strong emphasis on privacy.
“There’s one valid reason for [Apple] to take its time: if they genuinely regard privacy and data management as fundamental rights. And [Apple] does promote that kind of rhetoric,” Gruber notes.
“If they champion that as a top priority, they might encounter conflicting interests. By relaying all queries to OpenAI and providing the context required, they could likely enhance functionality, but doing so would compromise their privacy commitments.”
For years, a focus on privacy was seen as a reason Siri appeared less capable compared to, say, Google Assistant. It felt less intelligent and more artificial because it fundamentally had less information about you. Regardless of the accuracy of that perception, it lies at the heart of the challenges facing the new Siri as well.
A Tale of Two Parts
The new iteration of Siri is constructed around two primary components. A compact language model operates directly on the iPhone, while more sophisticated queries are redirected to OpenAI, with user permission required.
It’s estimated that Apple’s on-device AI systems encompass around 3 billion parameters, whereas estimates for the parameters in OpenAI’s GPT-4 reach approximately 1.8 trillion—about six hundred times the amount. DeepSeek gained attention as a more efficient, streamlined AI model in early 2025, yet it still boasts a reported 671 billion parameters.
