This isn’t a normal software delay. LLMs and Generative AI is not “normal” software in the sense of if you work harder on it, pour more money into it, give it more time, things will always get better.
There is a Computer Science gap that needs solving for before these things will ever work properly in the way they were demonstrated. Apple may be able to do that, but it will take a long time. This is why the summaries suck. Anyone who works in the field and isn’t also benefiting from AI hype financially or career-wise right now knows this technology has fundamental faults.
You can look up Yann LeCun‘s writing on these topics (not Apple specifically) who is the chief scientist of AI at Meta and has been working with neural networks since the 1980s. I’ve worked with dozens of CS PhDs and there are uses for this technology, but what we’re being sold as consumers is mostly not ready and there are profit motivations behind what’s being released, instead of solving for actual problems in a deterministic and concretely repeatable way.
There’s a public perception that because “computers always get faster” more and better hardware will solve a lot of the issues, but that just objectively not true. Without the transformer breakthrough in 2017 we wouldn’t have LLMs at all, maybe something akin to them but not LLMs as they currently exist. The most impressive thing Apple has done so far with AI in my opinion is create Private Cloud Compute, and that’s mostly on the hardware / firmware / big data side of things, not the actual algorithmic novel CS work that needs to leverage that compute in order to accurately produce results.
There’s a long, long way to go. Because Apple isn’t being led by someone who has total control, nuanced taste, and technical acumen, there is a large problem. Steve Jobs would have demanded to understand the technology before he got on stage to announce it, and either that didn’t happen with Apple’s current leadership, or they said “we’ll figure it out” and did it anyway. That’s a huge problem and people need to lose their jobs over it, not because it’s late but because there are surely engineers inside Apple who know those limitations, shared them, and weren’t listened to.
A lot of people hold Apple stock, or have it in their retirement portfolios and don’t want to publish things that can severely hurt the company. For a long time I thought Gruber was one of those people but either he divested or grew some confidence because it’s about time he called this debacle for what it is. It’s great reporitng. Kuo is just following-up on that saying “Apple is aware” but that really is not enough, to John’s point there needs to be a mobileMe style reckoning internally, and if it hasn’t happened then Apple is already rotten.
It may have happened to some extent by shifting Kim Vorrath, but we haven’t gotten any detailed reporting about that. And even though she may stop Apple from screwing up pre-announcements in the future, and holding developer’s feet to the fire internally, she can’t undo what was already shown, promised, and released in a fundamentally broken or underbaked state.