We're watching Apple fill the moat in.
Ability to download / update tiny models from Apple and Google as they improve, à la Google Maps.
No need for web services like ChatGPT.
This is more evidence that Apple is investing in building a MLLM as good as anything OpenAI and Google can build, albeit in a more Apple-y way (privacy-first, licensed content, etc.).
This will only work if their models (even their tiny or even medium / base models) equal (or are better than) GPT-3.5 / 4.
From there, Google will follow Apple in doing this offline / local LLM play with Gemini.
OpenAI's ChatGPT moat will certainly shrink a bit unless they release another powerful multimodal model.
If OpenAI wants to survive, they will need to expand way beyond their current business model of charging for access to an LLM. The logical place for them to go would be custom chipsets or ARM/RISCV IP blocks for inference.
Think of it in a physical sense. OpenAI is a high walled castle surrounded by a physical moat. This protects them and their business model. Apple comes along and builds a super tall tower right next to the moat. They can now see into OpenAI’s castle, fire arrows, catapult in a giant wooden badger, etc. Even if Open AI copies the design of Apple’s really tall tower and built it behind the moat and castle walls, it wouldn’t do much because Apple still would be able to get stuff over the moat and walls. The moat doesn’t matter anymore for the most part. The castle (OpenAI) can be compromised and needs bigger walls, relocating to someplace with a bigger, or a way of attacking the tower (Apple). Copying doesn’t really accomplish any of those three.
https://www.reuters.com/technology/chatgpt-owner-openai-is-e...
https://techcrunch.com/2023/09/27/openai-is-reportedly-in-ta...
As they say, the house always wins.
And they cannot bring software ecosystem for their hardware without Google, at least easily.