Apple rejected an astronomy app from the iOS App Store because “it was astrology”. When the author appealed, Apple doubled down because “We understand that the app includes a live tarot reading feature.” Ridiculous, right?
Well. The author in question is Terry Godier, and while I understand that he has gained quite the following in recent months, I am not a fan. He is unabashedly pro undisclosed AI writing, having conjured — “written” would be inappropriate in this context — a whole essay on large language models being a whole new communication technology. Trigger warning for those sensitive to LLM-ese.
The app, “Dark Hours”, was supposed to be an iOS-native version of this website. It has some truly useful information on the quality of the night sky for stargazing at any given location. It also has the tells of AI generation: too wordy — or overly verbose, if you prefer LLM talk — in some places, too technical in others. Many small digits scattered around the screen. It even manages to fit a this-not-that in what should be a technical explanation.
I have nothing against using LLMs as a coding aide. Dave Weiner sings Claude praises daily in his blog, while being realistic about its limits. Half of the ATP Podcast is an advertisement both paid and unpaid for OpenAI and Anthropic. The creator of Pedometer++, Widgetsmith and many other apps, “Underscore” David Smith, had a good post on his use cases. If you look at their products you could never tell that they were “made by” AI, because they weren’t. If any of them has ever made a fully “vibe-coded” app, they have yet to publish it under their name.
But then if you do fully vibe-code an app, and I am not a programmer so professionals please chime in, can you and do you vet every single line of code? “Live tarot reading” is a weird mistake for a human reviewing an app to make, unless there is a live tarot reading advertisement (or, why not, full functionality) somehow baked into the thousands of lines of code. Stranger things have happened than an app behaving differently depending on the IP address, though they usually try to hide something from the App Store reviewers, not show it to them.
All this to say, the story smells and I’d withhold judgement until we have more information. Unless, of course, it was Apple using automated LLM-based review even in the appeal process, in which case: a plague on all their houses.