Reading Between Apple's Open Source AI Lines...
Alex Kantrowitz
Last week I went on Alex Kantrowitz's Big Technology Podcast to talk all things Apple. But the majority of the conversation was around what they're likely to do with regard to AI. And given the OpenELM news yesterday, I think we nailed it.
While it had been all quiet on the LLM front for quite some time, Apple's release of eight (!) models of varying sizes to the open source community on Hugging Face (here's the white paper) would seem to indicate that this will be a big part of their approach to the space announced at WWDC in a few weeks. That is, Apple will focus their own AI efforts on things that can be done on device, while they will perhaps outsource other AI elements in the cloud to third parties.
It's a bit surprising that Apple released now as again, it would seem to telegraph their forthcoming moves, but this is also the way things are done in the AI community. And Apple has had to play ball in order to attract talent, if nothing else.1 As noted by Emilia David at The Verge:
In December, Apple launched MLX, a machine learning framework that ideally makes it easier for AI models to run better on Apple Silicon. It also released an image editing model called MGIE, which lets people fix photos with prompts. Another model, Ferret-UI, could be used to navigate smartphones. Apple is also rumored to be working on a code completion tool similar to GitHub’s Copilot.
We've talked about Ferret before (and Alex and I discussed on the pod), but there are likely to be a range of things Apple will aim to do on-device from an AI perspective. And I'd be shocked if we don't get an "aiKit" or whatnot.
Alex and I also discuss anything Apple might be doing in the robotics space. The general state of Apple's business, and why Services are so critical for them right now. And even a bit about Spyglass itself.
1 And it's not like Apple has no history with open source -- see: WebKit.