This market will resolve YES if Apple announces a generative AI model at WWDC 2024. The model must be developed by Apple, capable of creating new data (such as text or images) based on learned patterns, and should not involve the integration of an existing model, such as GPT-4, into their products.
The model must be generative in the sense that this word is commonly used. For example, the transformer-based enhanced autocomplete from WWDC 2023 would not count. In contrast, a tool that would allow users to rephrase an entire paragraph would.
1,000
3.00This market resolves YES. Although they didn’t explicitly say that they developed the models behind Apple Intelligence, the fact that the models run on-device shows that Apple has had control of the models, and specifically, the model behind the GenMoji feature must have at least been heavily modified from an existing model, sufficient to consider it a model developed by Apple.
@Predict Again, low confidence, but this feels very wrong to me.
the fact that the models run on-device shows that Apple has had control of the models
They 'had control' in the sense they chose which models to use, yes, but plenty of open source models run on apple silicon? This does not at all prove they 'developed' the models.
the model behind the GenMoji feature must have at least been heavily modified from an existing model
I am unsure 'modified' counts as developed. But that doesn't matter, because there have been open source emoji-generation models in the past, and I think all they'd need to do is figure out the right prompt for an existing image model and then downscale it? Why do you think it must've been "heavily modified"?
Also, weaker, even all that aside, the title says "announce their own". I don't think it's announcing their own generative model if they do not announce it but you have to infer it. It should at least require a higher burden of proof that I don't think the above is close to meeting
(if you decide you want to change the resolution another mod can do that for you it's perfectly normal and encouraged)
@jacksonpolack they did announce it, several times, "highly capable LLMs built into Apple Intelligence". Apple is not outsourcing that. They later specifically and separately mentioned what they are outsourcing, through the ChatGPT partnership/support, to OpenAI.
Apple is not outsourcing that
This is an unjustified statement that you are just making. I could just as easily assert that Apple is far behind in language modeling and is obviously going to outsource it to the best available OS models. I don't think we can conclude either.
they did announce it, several times, "highly capable LLMs built inbto Apple Intelligence
Yep, these could be open source, and those would not qualify for this market as they wouldn't be apple's "own" models and they would not have been "developed by apple".
They later specifically and separately mentioned what they are outsourcing, through the ChatGPT partnership/support, to OpenAI.
You're reversing the implication here? They are outsourcing ChatGPT, they might also be outsourcing other things, using external ChatGPT in no way implies they didn't also use OS models.
... How do we know any of the models involved were "developed by Apple"? I might have missed something but they could just be using off the shelf open-source models, which wouldn't be "models developed by apple".
A "model" in machine learning would refer, I think, to the blob of weights, not the product built on top of it
I currently think - low confidence but strong in weight - that this should resolve NO. It requires them to announce a model, and says "The model must be developed by Apple". I did not see a "model" announced anywhere". And that's debatable, but it's pretty clear that nowhere was it stated that a model was "developed by apple".
we probably won't get extra detail, but it seems very clear by "apple intelligence" and similar that it's an apple owned product.
> and should not involve the integration of an existing model, such as GPT-4, into their products.
despite the model involving an integration with gpt-4o, gpt-4o is only a very small part of the overall product, and surely would not make this market resolve NO; that would be absurd.
@deagol Probably not if it would be similar to just suggesting Siri Shortcuts. If the user can direct the generation of such automation then it probably would count.
It would also depend on the extent and complexity of these suggestions. If they are limited to just simple workflows with few steps and the user cannot ask for which scripts they want generated then it would almost certainly not count.
I assume generating code counts?
https://www.macrumors.com/2024/02/15/apple-xcode-ai-tool-for-coding/