The future will be local models. Let these investors burn the money.

Agree. What’s your reasoning?

Apple is already doing this:

https://www.apple.com/uk/apple-intelligence/

I wish, but why do you think so? Last decade the flow was from local compute to the server, not the other way around.

While building lix change control I figured out that collaboration was/is the driver behind local to cloud. A secondary reason is compute.

If a system exists that enables (distributed) collaboration on files, the appeal of cloud-based software vanishes. Remote compute becomes possible as well.

At least, that’s the bet we are taking with lix https://lix.opral.com/

I think a short-term driver towards local models is tooling. As difficult as reliably getting useful output can be, and as fragile as the resulting product can be, you want assurances of stability that don't exist with a model in someone else's cloud.

I'm not an expert on anything, though.

Yeah but you have to acknowledge the distance between local compute and a gazillion dollar data center, and even past that the hardware for a local setup is premium tier so it's not gonna quite catch up considering the cost.

i can’t imagine being a a trillion dollar company doing any serious business with a company where the entire executive suite has vacated except the ceo that’s clearly doing things.

at this phase, it would be disastrous for apple to bank their reputation on sam.

Is this actually related to the recent exodus? Apple is paying OpenAI nothing for use of their model on iPhones. Maybe they just realized being able to get away with paying nothing is a sign of how commoditized models will be.

At the time, I read the Apple deal as OA strategically finding an alternative to MS as the sugar-daddy, in terms of free cash flow and datacenter capabilities: it didn't need to be true, it just needed to be a compelling negotiation point with MS. The fact that Apple was outright investing in OA was in line with this view: if OA was just going to be some pluggable commodity, why invest in them? Apple doesn't do random VC for the hell of it.

So Apple withdrawing would seem to be bad news for OA, even if the rationale was something like "the valuation is now too high for our taste" rather than "we are losing faith in them as an organization" or "we think they are completely replaceable".

I am disappointment in the AI advancements in the recent years. It turns out progress is made not for the sake of the technology and the good for humankind, but for money. Just another pump and dump...

"It turns out?" "In recent years?" It has forever been so.

Is this the newest form of protest after Satya tried to burn the company to the ground? Throwing the toy away and saying "I don't want to play with this anymore, it's! It's a pump and dump"?

Apple is on track to own everything in the vertical except chip manufacturing. It would make sense for them to invest in their products and services over something with more varied interests.

They aren’t doing that either. Over 80Bn in cash, they’re doing their best Smaug impression.

Maybe Apple just realized that they would have bought the top of a massive pump and dump scheme with the huge risk of the researchers leaving the company whilst the training and inference costs suffocates OpenAI.

Sometimes, FOMOing into a pump and dump is not worth it and they were just too late.

I don’t think Apple was too late at all. In all the hype, nobody asks “how does this help $family_member in their day to day”.

I’ve used generative AI to help with writing documents and code. I think it’s fun for creating and modifying images. As someone with no creative talent, Photoshop has been quite fun to use.

But frankly, AI is completely useless in almost every way I see it implemented. I cannot see my grandparents, my mom, or my brother use it for anything practical. Everyone’s rushing to flesh out an investor slide deck without taking a moment to think: “but how can this actually be used in someone’s day-to-day”

I’ve enjoyed Apple’s extremely conservative approach so far — they’ve kept it mostly practical.

For example, let’s take the photos app. Just being able to search for “(dogs name) lake” and having images of my dog near a lake is day-to-day useful.

Another example is their introduction of writing tools to proofread blocks of text. My parents already use Spellcheck and Grammar check, this feels like the next step up.

Compare this to Samsung who took the SV Venture-Fund-appeal-style route and blasted “AI” everywhere they can call a function or put a button. Or Microsoft, who’s Copilot story is focused on imprinting keys on keyboards without answering “how does this help the end user”

This hype around AI is gaudy. Nearly every implementation feels cheap, half-baked, and obnoxious. If Apple’s skepticism of implementing AI so far is “late”, late is not a bad thing at all as an end user.

(All that said, Genmoji looks incredibly stupid and I’m wondering what they were thinking with this…)

I suspect we don't want it in our day to day, removing human interaction? But it is being used in day to regardless. Eg picking the next song to play.

It seems to excel at helping people do their jobs, double checking their work, etc. It's maddening since it makes constant mistakes. But Apple should be utilizing this to make HomeKit as magical as promised -- rather than automating email, which is basically the lowest hanging copy and paste able fruit available

You don’t see how an always available generative agent that could help monitor for cognitive decline and offer relief from loneliness could be helpful for the elderly?

Relief from loneliness? Maybe in the short term, but it sure seems like relying on a chat bot as a “friend” is likely to lead to even more isolation and loneliness long term, not more. Just take a look at the folks interacting with the “AI girlfriends”. It’s sad, really.

From my perspective,

> This hype around AI is gaudy. Nearly every implementation feels cheap, half-baked, and obnoxious.

is true but does not imply:

> AI is completely useless in almost every way I see it implemented

which is not true. It's just harder to get real value out of it than junk. And it's a completely new "skill" to get value out of LLMs which most of us aren't learning.

But why bother learning this as a skill? The time value simply does not exist despite the tech-bubble’s hype.

Let’s look at real-world usage. Two of the most grueling tasks most people encounter in a year are filing taxes and understanding contracts. LLMs can ingest a lot of data to help in both scenarios, but whatever it advises is useless until a company accepts liability for it’s output.

Can you say more about pump and dump?

If AMD can do it, they can as well: https://news.ycombinator.com/item?id=41674382

Apple realized that 1/OpenAI is no longer the ONLY game in town, and 2/Models != Platform. Maybe, maybe Apple can achieve what they need (Apple Intelligence), while keeping the doors open for multiple models.

> Apple realized that 1/OpenAI is no longer the ONLY game in town

Who is the other game in town for them?... Meta? Google?

No one from the list (also add Anthropic to that list) is SOTA right now, but maybe Apple wants to keep its options open as the gap closes between OAI and the others.

> No one from the list (also add Anthropic to that list) is SOTA right now

I don't use OpenAI's products. Is there an area where they are clearly ahead? I thought it was at least a tie in all categories.

Last time I checked, o1 did much better in PlanBench that models from other providers indicating better perf in that area [1][2]. Probably that lead will rapidly decay as other players implement similar approaches to "reasoning"/LRMs.

[1] https://www.arxiv.org/abs/2409.13373

[2] https://x.com/omarsar0/status/1838353480672563581

o1, which you seem to be referring to, is at best slightly better at specific tasks than Sonnet 3.5 while being completely unviable economically for production use, despite the amount of subsidizing OpenAI has been doing to make them lose a cool $5 billion this year. GPT has barely improved over the last 1.5 years [1].

[1] https://aider.chat/assets/models-over-time.svg

Yeah, I know. I didn't say it was incredibly better in every bench, but there are reasoning and planning benchmarks it does much better (not just slightly) than Sonnet 3.5 and the rest right now, and that still that makes their offering SOTA vs the others, even if its not in terms of cost.

I put sources in another sibling comment, and as I said in both comments, I think that gap will close. I don't expect OAI to keep their moat much longer, especially with the people they've lost.

PS: I say all of this as a constant critic of OAI: both how they behave and their offerings. But credit where credit is due, I don't see out-of-the-box alternatives to OAI's rlhf'd CoT at any price point right now.

Anthropic as well

My hunch is that OAI's models do not not really offer visible advantages over other open models for Apple's use cases. Even if OAI's models are more advanced, Apple can resort to cheaper post-processing to fill the gap. In Apple's scale, Apple should just host its own models to lower the long-term cost, as well as to have full control to iteratively improve the quality of the AI offerings.