In 2018, Craig Federighi (main vice-president of software engineering at Apple) announced it as the missing part of the puzzle: John Giannandrea, defector of Google, had to give a spine to artificial intelligence at Apple. Six years later, The site is not completed. The new Siri is only a whisper wave that is crazy, homemade language models are still not deployed and Apple Intelligence worries employees of the firm themselves. According to a senior member of the team quoted by Bloomberg, ” The ship has been taking the water for a long time ».
However, it is not for lack of interest. Steve Jobs himself had called the co-founder of Siri-Dag Kittlaus-24 days in a row to seal his takeover. However, the AI Apple has always been a halftone promise. Well present in speeches; much less in uses. Today, faced with the acceleration carried by Openai, Google or Microsoft, Apple tries to regain control, but seen from here, their tactics more like a series of contortions than a real action plan.
Apple Intelligence or Apple Illusion?
At WWDC 2024, Apple unveiled its late vision of artificial intelligence to the general public, under the name of Apple Intelligence. A set of functions supposed to transform in depth the user experience: Siri rethought, generation of automatic text, summaries of e-mails or notifications, personalized suggestions. On stage, everything seemed fluid, and fully mastered: The usual atmosphere of a keynote of this scale.
But if we believe the information of Bloombergthese demonstrations were in reality videos recorded from still unstable prototypes; None of the tools presented worked in real time. Rather than offering a technical response to its competitors, Apple therefore told us about a story.
No products found.
Why this Bérézina, for a company that had accustomed us to this equation: ” X Product ad = delivery of the X product in the following months »? Historically, Apple has always sold promises that materialize.
One of the obstacles is due to the very organization of teams : Siri, for example, remains distributed over several teams, with discordant visions. Mike Rockwell, who now oversees Siri engineering, first refused to postpone to Federighi, still skeptical a few years ago on the real utility of AI in iOS.
Internally, The cultivation of partitioning has slowed the changes expected more. The “LLM Siri” project, piloted from Zurich, must just unify the assistant around a conversational model. Another project, started much later than the competitors, who is not about to leave the laboratory.
No data, no miracle: the impasse of the Apple model
Apple has long seen AI with prudence, even distrust. A mined ground: too talkative, too invasive, too uncontrollable – everything that is detestable in his eyes. Giannandrea would have even argued, internally, that most users prefer to deactivate generative AI functions. The firm has therefore chosen a way other than competition: embed the local models, avoid remote servers, and respect a strict confidentiality framework. The problem is that this posture considerably limits the training of its models.
Make the choice not to collect massively data, It is to make an ethical choice, but also technological. This amounts, in practice, to accept a slower, less nourished, therefore less efficient training. As powerful as it is, an iPhone 16 Pro Max remains a personal device, not a cluster of supercalculators cooled to nitrogen.
With equal local power, on -board models simply cannot compete with those of Openai or Google Deepmind, watered every day by colossal volumes of requests, texts and human interactions.
The Crawler Applebot, for example, respects the exclusion beacons used by websites (robots.txt). Result: little external data usable. To compensate, Apple would have bet on synthetic training techniques by locally analyzing email fragments, without these data leaving the device. A choice consistent with Apple philosophy, but which, technically, works like a hand brake pulled permanently. Impossible to reach the speed of AI giants when We refuse to use the same data highways.
The firm multiplies the tracks to stay afloat. Integration of Chatgptdiscussions with Perplexity AI, development of an internal chatbot deemed ” equivalent to GPT-4 According to executives. The idea would be to strengthen Siri and prepare for a possible breach of the lucrative contract which makes Google the default engine of safari: an agreement now in the crosshairs of the regulatory authorities.
At Cupertino, AI has become a subject too visible to be ignored, but still far too sensitive To be fully assumed. Hence this clumsy gymnastics we are witnessing for almost a year : Show without exposing too much, to announce with great fanfare without delivering anything to us and repel by maintaining communication. And now, All this has a name: Apple Intelligence.
- Apple still struggles to make its ambitions in artificial intelligence, despite repeated announcements and carefully staged demonstrations.
- Internally, the project is slowed down by organizational tensions, a fragmented strategy and a cautious approach to data.
- Result: the firm advances in a setback against its competitors, trapped in its closed model and its culture of secrecy.
By: keleops ag
