For the next big Twitter-killer social media network, our trademark bad idea is that we're not gonna let users enter alt-text:
Instead we're just using an LLM to identify what's in the photo.
I suppose that given the absence of consideration for content licensing here, someone will be training ML on our goodwill alt-texts, if they're not already doing it.
The real killer feature is: You won’t have to type any of your own posts, we auto-post for you based on a model we trained on your private messages with friends and family.
That's just phase one.
Phase two we protect your identity by taking those descriptions and feeding them into an image-generation AI. So you can't be identified from your selfies with your cat, as you're not technically in them