Apple Intelligence-Related Instructions ⇥ theverge.com

Reddit user devanxd2000:

I was digging into the system files for the update and I found a bunch of json files containing what appears to be prompts given to the AI in the background. I found it interesting and thought I’d share.

You can find them here: /System/Library/AssetsV2/​com_apple_​MobileAsset_UAF_FM​_GenerativeModels

There’ll be a bunch of folders, some of them will have metadata.json files like this.

Wes Davis, the Verge:

Files I browsed through refer to the model as “ajax,” which some Verge readers might recall as the rumored internal name for Apple’s LLM last year.

It is unclear to me if these directly represent the instructions which interpret and produce the results users see. These could be something else, like a file involved in the development process but not related to how it functions on a user’s device; we just do not know.

But, assuming — quite fairly, I might add — that these instructions are what underpins features like message summaries and custom Memories in Photos, it is kind of interesting to see them written in plain English. They advise the model to “only output valid [JSON] and nothing else”, and warn it “do not hallucinate” and “do not make up factual information”. The latter two are just good rules for life. I am not sure what I expected, but I guess it was not these kinds of visible instructions. But, I guess it would make sense for it to feed through what I presume is the same system underpinning the revised version of Siri, which needs to interpret everything from plain English commands. After all, programming is just a specific version of a language.