Apple’s AI Reboot: Can ‘LLM Siri’ Finally Get It Right?
Apple’s AI Reboot: Can ‘LLM Siri’ Finally Get It Right?
Apple’s foray into the world of Artificial Intelligence, particularly with its digital assistant Siri and the broader suite of features dubbed “Apple Intelligence,” hasn’t exactly been a smooth ride. Since its initial rollout, Apple Intelligence has faced numerous challenges. Now, a detailed report from Bloomberg’s Mark Gurman sheds light on the reasons behind these struggles and, more importantly, how Apple is attempting to course-correct, primarily by rebuilding Siri from the ground up.
This isn’t just a minor tweak; it’s a fundamental rethinking of how Siri operates, moving away from its current architecture towards a generative AI-powered system internally known as ‘LLM Siri.’ But why the need for such a drastic overhaul? Let’s delve into the issues and explore Apple’s ambitious plans for the future of AI.
The Siri Struggle: A Deep Dive into What Went Wrong
For years, Siri has been the subject of jokes and criticisms, often lagging behind competitors like Google Assistant and Amazon’s Alexa in terms of accuracy, comprehension, and overall usefulness. The attempt to simply bolt on generative AI features to the existing Siri infrastructure proved to be more problematic than initially anticipated. According to Gurman’s report, several factors contributed to Apple’s AI woes:
1. Hesitant Investment in AI
According to Gurman’s sources, Apple’s software chief, Craig Federighi, was initially hesitant to make substantial investments in AI. Apple typically prefers to invest in projects with clear, defined goals. However, in the rapidly evolving field of AI, the exact endpoint is often unclear until significant development has taken place. As one unnamed Apple executive pointed out, “…you really don’t know what the product is until you’ve done the investment.”
This cautious approach meant that Apple was slow to acquire the expensive GPUs necessary for training large AI models, putting them at a disadvantage compared to competitors who aggressively invested in the technology.
2. Late to the AI Game
Another contributing factor was Apple’s late entry into the AI race. One executive revealed that Apple Intelligence “wasn’t even an idea” before the launch of ChatGPT in late 2022. This late start meant that Apple had to play catch-up, scrambling to develop AI capabilities while other companies had already made significant progress.
3. Misjudging User Desire for AI Chatbots
John Giannandrea, Apple’s AI chief, reportedly believed that users were not particularly interested in AI chatbots, citing the common desire to disable tools like ChatGPT. This perspective may have influenced Apple’s initial approach to AI, leading them to underestimate the potential of generative AI and its impact on user expectations.
4. Incompatible Architectures: Old Siri vs. New AI
Apple’s initial strategy of integrating generative AI features with the existing Siri architecture proved to be a major stumbling block. The two systems simply didn’t mesh well, leading to a constant stream of problems. As one employee described it, “It’s whack-a-mole. You fix one issue, and three more crop up.” This suggests that the underlying architecture of the original Siri was fundamentally incompatible with the new AI capabilities.
5. Leadership and Internal Dynamics
Giannandrea’s role within Apple’s leadership structure also played a part. As an outside executive hire, he reportedly didn’t have the same influence or “forceful” personality as other key figures within the company. This may have hindered his ability to secure the necessary funding and resources for AI development. Furthermore, some employees felt that Giannandrea didn’t push his team hard enough and underestimated the threat posed by AI giants like OpenAI and Google.
6. Overzealous Marketing
Apple’s marketing efforts may have also contributed to the perception of AI shortcomings. The company heavily promoted promised features, such as an improved Siri and Apple Intelligence’s ability to draw context from apps across the system. However, these features were not fully ready, leading to delays and disappointment among users.
The Path Forward: Rebuilding Siri from the Ground Up
Recognizing the shortcomings of its initial approach, Apple is now embarking on a significant overhaul of its AI strategy. A key component of this effort is the complete rebuilding of Siri, moving away from the existing architecture and embracing a new, LLM-based engine.
According to Gurman, Apple’s AI team in Zurich is working on this new architecture, which will be entirely based on a Large Language Model. This approach aims to make Siri more conversational and better at synthesizing information, ultimately providing a more natural and intuitive user experience. This project aims to make Siri “more believably conversational and better at synthesizing information.”
Leveraging iPhones and Differential Privacy
In addition to the architectural overhaul, Apple is exploring innovative ways to improve its AI models while prioritizing user privacy. One method involves leveraging iPhones and differential privacy to enhance Apple’s synthesized data. This process involves comparing fake training data with language from iPhone users’ emails, but doing so on-device and sending only the synthesized data back to Apple for AI training. This approach allows Apple to improve its AI models without compromising user privacy.
Siri as an AI Web Search Tool
Another intriguing possibility is transforming Siri into an AI-powered web search tool, similar to Perplexity. This would involve allowing the LLM version of Siri to access and synthesize data from multiple sources across the web. Apple has even reportedly approached Perplexity about a potential partnership for AI search in Safari, indicating a strong interest in this direction.
Giannandrea’s Changing Role
Interestingly, John Giannandrea has reportedly been removed from direct product development, Siri, and robotics projects. While Apple executives have considered putting him “on a path to retirement,” they are concerned about the potential departure of key research and engineering personnel who joined Apple under his leadership. Regardless of his future role, Giannandrea is reportedly “relieved Siri is now someone else’s problem.”
Conclusion: A Make-or-Break Moment for Apple Intelligence
Apple’s AI ambitions face a critical juncture. The company’s initial struggles with Siri and Apple Intelligence have highlighted the challenges of entering the AI race late and attempting to integrate new technologies with existing infrastructure. The decision to rebuild Siri from the ground up, leveraging LLMs and prioritizing user privacy, represents a bold attempt to regain lost ground.
Whether Apple can successfully execute this ambitious plan remains to be seen. The success of ‘LLM Siri’ will be crucial in determining the future of Apple Intelligence and its ability to compete with other leading AI platforms. The coming months will be pivotal as Apple strives to deliver on its promise of a smarter, more capable, and truly helpful digital assistant.
Source: The Verge
