blog: apple image playground nazi cat salute blog post
This commit is contained in:
parent
a7f21115ac
commit
0741a8a80a
1 changed files with 40 additions and 0 deletions
|
|
@ -0,0 +1,40 @@
|
|||
---
|
||||
title: Apple Intelligence made me a cat doing a Nazi salute (?)
|
||||
date: 2025-03-10T22:52:12.082Z
|
||||
slug: 2025-03-10-apple-intelligence-made-me-a-cat-doing-a-nazi-salute
|
||||
author: Thomas Wilson
|
||||
|
||||
---
|
||||
When Apple [announced their suite of Apple Intelligence features](https://www.apple.com/newsroom/2024/06/introducing-apple-intelligence-for-iphone-ipad-and-mac/) in 2024, they did so nearly two whole years after OpenAI announced both [DALL·E](https://openai.com/index/dall-e-now-available-without-waitlist/), their image generation tool, and [ChatGPT](https://openai.com/index/chatgpt/), their chat-like interface to a Large Language Model.
|
||||
|
||||
For whatever reason, these later tools were the starting pistol to the "can this be AI?" race we find ourselves in. It's okay to take one's time to build something you're proud of, and sure of. I think Apple's hand was forced by a stock market that watched company after company reveal that actually they're all also AI companies: Meta [releasing Llama](https://arxiv.org/abs/2302.13971) in February 2023 , Google [announcing Gemini](https://blog.google/technology/ai/google-gemini-ai/) in December 2023, Amazon taking a long run up to a [Alexa Plus](https://www.amazon.com/dp/B0DCCNHWV5) in March 2025.
|
||||
|
||||
Apple gave Apple Intelligence [the closing forty minutes](https://www.youtube.com/live/RXeOiIDNNek?si=OaI6KNxMsdK8PKcN&t=3871) of their June 2024 keynote, promising various generative and utility AI features.
|
||||
|
||||
One of those features was Image Playgrounds: Apple's play at image generation *a la* DALL·E or literally countless other examples if you just type `"AI image generation"` into anything that looks like a search engine. You'll find one, I promise.
|
||||
|
||||
The problem is, very famously, these image generation tools are sorta janky. And also, you know what kinds of things can be signalled or identified visually: parts of race, gender, and religion. You wanna let a little random machine, trained with images on the open internet[^Some popular tools likely included Child Sexual Abuse Material (CSAM) images in their training source too ([source](https://purl.stanford.edu/kh752sm9123)). I doubt, strongly, that this is the case for Apple, but finding proof of a negative is famously hard.] , generate an image with ideas of race, gender, and religion - and then put your brand name next to it ?
|
||||
|
||||
You might remember that Google [apologised](https://www.theverge.com/2024/2/21/24079371/google-ai-gemini-generative-inaccurate-historical) after their generative AI tool insisted on generating racially diverse American Founding Fathers.
|
||||
|
||||
Anyway, my first experience with Apple Intelligence Image Playground generated me a cat with two tails and also doing something that looks too much like a Nazi salute (with its fifth arm?):
|
||||
|
||||

|
||||
|
||||
You know what I asked for ? `"A black cat, stretching"`. It generated me some uninteresting, and non-dynamic, obviously AI-generated images of black cats in a vaguely stretch-like position:
|
||||
|
||||

|
||||
|
||||
When I was asked if I wanted to modify my image, I said yes! `"a witch's cabin in the woods"`. And I received the image of the cat with it's fifth arm raised.
|
||||
|
||||
I quickly saved the image, and now the app appears to have forgotten all about it. I opened it today, and asked for a `"black cat doing a stretch"` and got the above image. When I promote it with `"a witch's cabin in the woods"`, I got a cat in the cabin, but no longer stretching:
|
||||
|
||||

|
||||
|
||||
For love, nor money, I cannot get Apple to generate something similar, even if I prompt it with: `"doing a big stretch"` and `"with it's arms up"`
|
||||
|
||||
This was my *first* experience with Image Playgrounds. Even if I didn't have complicated feelings about generative AI and human creativity (I sorta think image generation is gross, I rarely think it's twee or cute as the companies try to pitch it; but also I used it on my own wedding website), I don't love the relatively high chances of accidentally generating an image with hate-speech symbolism.
|
||||
|
||||
It seems as though something has changed internally which has made this exact image much harder (but probably not impossible) to generate. And that's another one of the problems: moderation of and constraining the models themselves.
|
||||
|
||||
I don't know. *Imagining* a witch's cat doing a big stretch brings me so much internal joy. None of the above images really make me feel anything much. Maybe it's because I know they're AI-generated, or maybe the air is too thin up here on my moral high ground.
|
||||
Loading…
Reference in a new issue