
The initiative, internally dubbed "LLM Siri," aims to integrate large language models to make Siri more conversational and context-aware. However, the project has been marred by technical difficulties and shifting strategies. Initially, Apple considered deploying both small and large language models—codenamed "Mini Mouse" and "Mighty Mouse"—to handle on-device and cloud-based processing, respectively. This approach was later abandoned in favor of a unified large language model, leading to frustration among engineers and the departure of key staff members.
Compounding the issue, Apple's AI division has been criticized for its indecisive leadership and risk-averse culture. Former employees have described the group as "AIMLess," citing a lack of clear direction and ambition. The Siri project has reportedly been passed between teams without substantial progress, highlighting systemic issues within the company's AI strategy.
The challenges have prompted a reshuffling of leadership. John Giannandrea, who joined Apple from Google in 2018 to spearhead AI efforts, has been moved out of product development roles. Craig Federighi, Apple's Senior Vice President of Software Engineering, has taken a more hands-on approach, encouraging the use of open-source models and external partnerships to accelerate development.
Despite these efforts, Apple's AI initiatives remain behind competitors. Internal studies have shown that OpenAI's ChatGPT outperforms Siri by 25% in accuracy and can handle 30% more queries. The company had planned to roll out enhanced Siri features with the iOS 18.4 update, but persistent bugs and underwhelming performance have pushed the release to at least 2026.
At the 2024 Worldwide Developers Conference, Apple introduced "Apple Intelligence," a suite of AI-powered features designed to enhance user experience while maintaining privacy. The announcement included a partnership with OpenAI to integrate ChatGPT into Siri. However, the demonstration raised eyebrows among Apple employees, as many of the showcased features were not yet functional on test devices.
Topics
Technology