Sources: many people inside Apple do not believe its own AI/ML team can deliver on generative AI, and worry that the company will only use that team's output
Speaking of Mark Gurman, he has an interesting bit about Apple's AI efforts in his latest Power On column:
Daring FireballJohn Gruber
Context & Ripple Effects
This report captures an early internal-confidence problem in Apple's generative-AI effort: the concern was not only execution, but whether the company would constrain itself to a single internal source of models and capabilities.
Later coverage describes Apple making generative AI a tent-pole project after executives tested ChatGPT, while subsequent reporting said some employees viewed the company as materially behind leading rivals in generative AI. Together, those reports make the earlier doubts relevant as a signal of organizational and technical readiness rather than a one-off personnel complaint.
First-order effects
Reported skepticism inside Apple’s AI/ML organization can weaken confidence in internally developed generative-AI output and raise the stakes for the team’s delivery.
If Apple relies only on that team’s work, product groups have fewer alternative paths when internal models or features do not meet expectations.
Second-order effects
Pressure to close an execution gap can push Apple toward a broader architecture spanning local and cloud processing, an approach later associated with Project Greymatter.
The report strengthens the case for evaluating outside model partners alongside internal development, rather than treating in-house output as the sole route to generative-AI features.
Third-order effects
If this pattern persists, generative-AI competition will be shaped as much by a company’s ability to integrate external capabilities and coordinate product teams as by ownership of a frontier model.
For Apple, a durable gap between internal ambition and delivery could make its distribution reach more dependent on partners, while preserving control over the user experience as the central strategic constraint.
The trend: This is an early data point in the shift from treating generative AI as a self-contained research capability to treating it as a product-integration and sourcing decision.
You are right in that large language models really caught Apple off guard, but at the same time it's not like they haven't been making moves in this direction. All of their M model chips have neural chips built into them that are tailor made to handle AI specific tasks on-device…
Putting it out there that Apple has missed the boat developing generative AI for its products. Seems like Meta, Microsoft and others are well ahead of the curve.
I'm sure Apple have a bunch of people on it - and you're right, it's not like they wouldn't have seen all this coming down the pike. My hope is that they can either improve Siri wholesale, or integrate the AI modelling into Siri to make it more intuitive. …
Disagree heavily. Apple is known not to be a first-mover. They also are known to invest in user experience first and foremost. They will utilize AI models in small use cases, like autocomplete, and then eventually add generative AI when hallucinations can be reduced. They hav…
Siri and Alexa have found niche use cases as timers and music players after about a decade of existence. Controlling peripherals like lights involves home automation integration that most people have not and will not setup. They are not generally useful after years of investmen…
I read through the replies and I believe that people completely miss the point here: we need better conversational assistant, it has nothing to do with Apple ecosystem, that can be easily plugged into any chatbot like ChatGPT or Google Bard. The key Apple's problem is inability …
My understanding is that Siri has always sucked because Apple insists on doing more work locally on the device relative to Google and MS. The interesting thing though is we're maybe only a couple years away from being able to efficiently run LLMs on mobile devices. Curious if th…
Siri and Alexa are more casual. Practical stuff like “What NFL teams are playing tonight?” or “Is it going to rain tomorrow?” or “Set a timer for 10 minutes.” That's very different from ChatGPT, which is more powerful but doesn't integrate with your ecosystem of apps. …
I will believe Apple is taking generative AI seriously when they delete all of the Siri code and replace it with a wrapper over ChatGPT. Siri has been a failed experiment for 12 years and it's time put it out of our misery for the good of every iPhone user.
The successful part of voice assistants is the hands-free command part. Turn on my lights, start a timer, play a song. I don't think improving the tech that enables them to have a conversation or answer questions would convince me to use a voice assistant for those purposes.
Apple is 3-5 years behind in AI overall & 2-3 years behind in GenAI (which is new). Impediments include Apple's secrecy culture which turns away top researchers and product builders and a hardware first approach which plays to their main business but slows them down. …
Wow. Great scoop by @StephenNellis and @maxcherney. The end of “Wintel” would mark the end of one of the tech industry's most important chapters (and be devastating to Intel). Exclusive: Nvidia to make Arm-based PC chips in major new challenge to Intel https://www.reuters.com/...