4 ms·

Why is Apple struggling so hard with LLMs? I get that they're a hardware company, but surely that can't be the full story. They must be suffering from some case

Why is Apple struggling so hard with LLMs? I get that they're a hardware company, but surely that can't be the full story. They must be suffering from some case of extreme perfectionism or something like that. I would assume it'd be very difficult to get an LLM to perform consistently and reliably without making it useless.

helveticabold48 2y ago
My guess: Their privacy centric image/strategy is a two-edged sword. You can only do so much on device locally.
- LoganDark 2y ago
  This has already been solved since last year with Private Cloud Compute.
  - jsemrau 2y ago
    Great point
  - saagarjha 2y ago
    Assuming they are actually ready, that is.
    - LoganDark 2y ago
      Private Cloud Compute is already used for existing Apple Intelligence features (all, like, one total of them).
- resource_waste 2y ago
  Its wild to me that people actually believe the privacy marketing. I guess not... there is a reason Apple is still in business, they are great at deception. It just seems so silly to believe a blackbox company that was outed by Snowden in PRISM would actually be privacy focused. Or that a company that touts security has been hacked so easily by Pegaus that anyone even buys into the marketing. But hey, people are religious, so I should be less surprised that people trust others.
mvdtnz 2y ago
> They must be suffering from some case of extreme perfectionism or something like that. Are you serious? Did you miss the headlines around their news and text message summary products?
- LoganDark 2y ago
  > Are you serious? Did you miss the headlines around their news and text message summary products? Actually, riddle me this: Do you really think they would just ignore how that went?
swatcoder 2y ago
No commercialized LLM product released so far would meet Steve Jobs' standard for a high-profile Apple product, and while that standard is clearly much more lax now, there still is one and still higher than what people accept from OpenAI et al. I wouldn't call that perfectionism, just a struggling brand standard that can't afford to lose even more face. I think many inside and outside Apple hoped that the ways that they scoped their features in the Apple Intelligence announcement would help them pull off something duly reliable and practical, but it's not that surprising that even those ambitions might have bought too deeply into the hype.
- LoganDark 2y ago
  > I wouldn't call that perfectionism, just a struggling brand standard that can't afford to lose even more face. It's very well-known that Apple is perfectionistic. I'm not meaning to say that perfectionism is a negative quality or a bad thing, just that it takes a while.
  - TheOtherHobbes 2y ago
    Apple is not perfectionistic. Apple is performative. The entire company is performing software development instead of actually doing it. Apple's development process is a marketing pitch-driven hallucination - project management by buzzword and individual career status progression. It's almost entirely inward-looking. The connection to Rest of World is increasingly mythical and remote. Some good work gets done in spite of this. But senior management doesn't understand quality - either in the internal sense of having bug-free robust product, or the design sense, where products meet real user needs in a satisfying, creative, and delightful way. Nice graphic design though. Apple is still the leader there. Processor dev has also been exceptional. IMO it's time for most of the C-suite to step down and let much younger talent take over and shake things up.
    - rightbyte 2y ago
      Has things gone cracy since I last used OSX for real in 2008 or whatever? Windows have become such a shitshow since Windows 7. I kinda assumed Apple didn't follow suite.
petesergeant 2y ago
> Why is Apple struggling so hard with LLMs? Depends what you mean. Building an LLM seems to be easy enough that several companies have built one, and even Twitter was able to magic one out of the ether from pretty much a standing start given the resources. Getting an LLM to do the right thing for you 99.999% of the time? Has anyone managed that yet? Frontier models still hallucinate the most basic shit, and consistency is very, very difficult.
rsynnott 2y ago
Theory: a lot of this work was started a few years back, when it was a common belief that LLMs would continue to improve at a breakneck pace. That has now happened, so now they’re stuck with a lot of clearly Not Good Enough stuff.
xvector 2y ago
They refuse to pony up for competent talent in ML. Meta, OpenAI, and others out-compensate Apple for ML talent by a factor of 2-4x.
- margorczynski 2y ago
  From what I'm seeing all those companies now have problems explaining to investors when they'll see any returns on all that investment. Especially after China has entered the game and DeepSeek disintegrating most of the moat that existed around those companies. And Apple can simply use what others created instead of burning billions into R&D which might not come up with anything noteworthy.
tippytippytango 2y ago
People expect the llm version of siri to be an agent, not a chat bot. There's a very wide gulf between a chatbot like chatgpt and an agent like... well I haven't found any good agents I'd give write access to my calendar yet so I don't know what to cite.

helveticabold48 2y ago

My guess: Their privacy centric image/strategy is a two-edged sword. You can only do so much on device locally.

LoganDark 2y ago
This has already been solved since last year with Private Cloud Compute.
- jsemrau 2y ago
  Great point
- saagarjha 2y ago
  Assuming they are actually ready, that is.
  - LoganDark 2y ago
    Private Cloud Compute is already used for existing Apple Intelligence features (all, like, one total of them).
resource_waste 2y ago
Its wild to me that people actually believe the privacy marketing. I guess not... there is a reason Apple is still in business, they are great at deception. It just seems so silly to believe a blackbox company that was outed by Snowden in PRISM would actually be privacy focused. Or that a company that touts security has been hacked so easily by Pegaus that anyone even buys into the marketing. But hey, people are religious, so I should be less surprised that people trust others.

mvdtnz 2y ago

> They must be suffering from some case of extreme perfectionism or something like that. Are you serious? Did you miss the headlines around their news and text message summary products?

LoganDark 2y ago
> Are you serious? Did you miss the headlines around their news and text message summary products? Actually, riddle me this: Do you really think they would just ignore how that went?

swatcoder 2y ago

No commercialized LLM product released so far would meet Steve Jobs' standard for a high-profile Apple product, and while that standard is clearly much more lax now, there still is one and still higher than what people accept from OpenAI et al. I wouldn't call that perfectionism, just a struggling brand standard that can't afford to lose even more face. I think many inside and outside Apple hoped that the ways that they scoped their features in the Apple Intelligence announcement would help them pull off something duly reliable and practical, but it's not that surprising that even those ambitions might have bought too deeply into the hype.

LoganDark 2y ago
> I wouldn't call that perfectionism, just a struggling brand standard that can't afford to lose even more face. It's very well-known that Apple is perfectionistic. I'm not meaning to say that perfectionism is a negative quality or a bad thing, just that it takes a while.
- TheOtherHobbes 2y ago
  Apple is not perfectionistic. Apple is performative. The entire company is performing software development instead of actually doing it. Apple's development process is a marketing pitch-driven hallucination - project management by buzzword and individual career status progression. It's almost entirely inward-looking. The connection to Rest of World is increasingly mythical and remote. Some good work gets done in spite of this. But senior management doesn't understand quality - either in the internal sense of having bug-free robust product, or the design sense, where products meet real user needs in a satisfying, creative, and delightful way. Nice graphic design though. Apple is still the leader there. Processor dev has also been exceptional. IMO it's time for most of the C-suite to step down and let much younger talent take over and shake things up.
  - rightbyte 2y ago
    Has things gone cracy since I last used OSX for real in 2008 or whatever? Windows have become such a shitshow since Windows 7. I kinda assumed Apple didn't follow suite.

petesergeant 2y ago

> Why is Apple struggling so hard with LLMs? Depends what you mean. Building an LLM seems to be easy enough that several companies have built one, and even Twitter was able to magic one out of the ether from pretty much a standing start given the resources. Getting an LLM to do the right thing for you 99.999% of the time? Has anyone managed that yet? Frontier models still hallucinate the most basic shit, and consistency is very, very difficult.

rsynnott 2y ago

Theory: a lot of this work was started a few years back, when it was a common belief that LLMs would continue to improve at a breakneck pace. That has now happened, so now they’re stuck with a lot of clearly Not Good Enough stuff.

xvector 2y ago

They refuse to pony up for competent talent in ML. Meta, OpenAI, and others out-compensate Apple for ML talent by a factor of 2-4x.

margorczynski 2y ago
From what I'm seeing all those companies now have problems explaining to investors when they'll see any returns on all that investment. Especially after China has entered the game and DeepSeek disintegrating most of the moat that existed around those companies. And Apple can simply use what others created instead of burning billions into R&D which might not come up with anything noteworthy.

tippytippytango 2y ago

People expect the llm version of siri to be an agent, not a chat bot. There's a very wide gulf between a chatbot like chatgpt and an agent like... well I haven't found any good agents I'd give write access to my calendar yet so I don't know what to cite.