Our culture is AI generated at this point

tldr : Every major social media feed uses transformers, the same technology as ChatGPT, to decide what post to show you. This is true for billions of people. What trends on social media decides our culture. This is the culture you exist in, even if you don't use any AI generated feeds.

Scrolling on social media is as bad as spending all day talking to ChatGPT. It is the same thing.

Apps with AI generated feeds include X, BlueSky (Discover), TikTok, Facebook, Instagram, YouTube, Spotify, Netflix, and Pinterest.

We know this because they've published how they've used transformers to create more engaging algorithms.

I'm not going to write a long post here. I'll just post a timeline. The longer version of this post includes the bastard Ronald Reagan, believe it or not.

• 1990s? The promise of "big data" is that "big computers" will eventually come and let you use all that data. Some of it today, the rest tomorrow. You just gotta invest in more storage, I swear, something good will come out of it.

• 2011: A neural network, once the underdog loser of "artificial intelligence", wins an image recognition competition by virtue of being implemented on a GPU. Loads of money starts to get poured into neural network research. IMO this is when the modern "AI era" really starts. "Coq lovers" are dismayed.

• 2017: "Attention is All You Need" is published, giving us the transformer architecture for "seq2seq" / "txt2txt". We'd understand this as "generative AI". This improved on generative adversarial networks (GANs) and variational autoencoders (VAEs) and capsule networks (haha hinton sorry man, attention really is all you need).

  • Side note: Neural network architectures are really interesting, because they are discrete concepts (rather than contiguous ones), but are so fine grained enough that it's hard to say where one architecture stops and another starts. Despite that, we have so many names for specific architectures. All this is to say is, please don't "uhm actually" me on the specifics of attention / bert / transformers. it's complicated!

    • Side side note: This will come up later.

• 2019: "ChatGPT 2" comes out, and millenials are like "Wow, this is way better than Cleverbot at being a chatbot. Cool! I'm sure this won't become the most reviled thing." But very overfit, sometimes generating exact copies of IRC and SS13 logs in what I could find.

• The exploding years : People aren't thinking about "AI" yet, but around this time, tons of apps come out with are utilizing transformer based recommendations, driven in no small part by Douyin's introduction to the US market as TikTok. Just as everyone got finished copying Snapchat's stories, now they had to figure out how to keep up with TikTok's engagement metrics. They do that by spending a lot of time figuring out how to use the Cool And New Neural Networks to do it.

• 2022: "ChatGPT 3" comes out, and "AI" as people understand it explodes. You know the rest since you are very online.

At this point, tons of big social media sites have published the ways they're using Transformers to improve search recommendations. And to be clear, these are AI generated feeds. All the flaws of the Algorithm are combined with the probably-brain-destroying capacities of these neural networks.

If you don't like AI, here is one social media site with absolutely no AI recommendation algorithm going on in the background.