There is some speculation online that Apple might end up having its own LLM
Criteria
Resolves positively iff...
Before December 31st 2023, at midnight, Apple announces its own large language model running on Apple devices
For the sake of resolving this market, I will define the "Large" in LLM to mean >=1B parameters
Said LLM may have been co-developed in an external collaboration, but is proprietary to Apple, e.g. it doesn't count if they just integrate an open-source model into some SDK, or announce a deep integration of an OpenAI GPT API or similar
Feel free to discuss if you think there's important criteria to be added.
Related questions
🏅 Top traders
| # | Name | Total profit |
|---|---|---|
| 1 | Ṁ531 | |
| 2 | Ṁ403 | |
| 3 | Ṁ328 | |
| 4 | Ṁ277 | |
| 5 | Ṁ260 |
@esusatyo afaik the model should be able to run on macs (only 7/13b) so i think it meets resolution criteria
@deagol i interpreted the second point as being “apple-developed” which this is, but not necessarily closed source? it’s not just a rehashed llava or smth
@deagol i think the prev paper was abt running a open source model developed by a third party faster, but this seems to be a fresh model
@ashly_webb some links might be helpful for @prodict to assess if this “is running on Apple devices” for normal Apple customers (not just devs) to take advantage of in any way, and if the non-commercial license fits the spirit of the question.
IMO this is just AI research, not an integrated product yet, as implied by the quoted tweet in the description, and the trading in this market backs this. Note the ferret paper came out in Oct and was widely reported in early Nov, yet not a blip in this market.
@deagol hmm, maybe i’m misinterpreting the resolution criteria. i didn’t realize it meant that the llm should be a released product. probably going to sell my shares
@ashly_webb @deagol @esusatyo I thought this through a bit. I realize the "proprietary" in the resolution criteria may have been ambiguous. I haven't seen this kind of outcome coming, but a) I think the spirit of the market was even more "integrated/released product" than "research model" than the criteria stubbornly interpreted would insist upon, b) this is CC-BY-NC which isn't considered entirely "free", but I think does not fit common usage of the term "proprietary" (see e.g. https://en.wikipedia.org/wiki/Proprietary_software#Types)
All things considered I've decided to leave this down as a "No" as of today -- where the market currently stands seems to back up this interpretation of the criteria/spirit too.
Does this academic paper count? It’s authored by Apple.
https://x.com/gabrielnocode/status/1737307286887133552?s=46&t=TuoHniragWyc8BOTqtjucw
@esusatyo Direct link to the PDF of the paper that clearly shows Apple as the author.
@esusatyo while authored by people working at AAPL, this looks like an inference method demonstrated with existing open-source LLMs (Falcon by TII, OPT by Meta), not an announcement or demonstration of a proprietary LLM by Apple, so doesn't qualify
someone dug deep into the Ventura Sonoma beta and estimates about 34 million parameters. So, I guess it means NO.
Technically you can already use an LLM on some of their Siri platforms. The jokes you can request are generated by Siri Natural Language Generation on tvOS 16.4+ beta: https://9to5mac.com/2023/03/16/apple-testing-siri-natural-language-generating-chatgpt/
But, yes, that's not "announced". Maybe tvOS 17 will have an extended version of their natural language processing system:
@esusatyo As long as it runs on-device and fits the criteria in the description, I don't think it should matter if it has its own user-facing app/"frontend" or is something developers can use inside their apps. And if it runs on-device, I don't think the kind of app apple integrates it in should matter, so "smart siri" and "on-device copilot in xcode" would both count IMO
Starting a related market:
https://manifold.markets/TravisLuckenbaugh/will-apple-announce-copilotlike-fun