A brief prediction on the future of generative models
As it stands, the generative models colloquially known as “AI Slop”, the way they gather data and operate are highly dubious:
- Most of the software and data sets are closed source, yet its progress is standing on the shoulders of everything that is open source.
- Feeding the computation power that enables machine learning is very resource intensive. It brings serious environmental impact and provides very little for people who have high standards for information. 1 2
- Vacuuming data from the internet as feed for these machine learning operations is also very resource intensive. Scrapers are hammering servers all around the globe, not respecting the policies set by system administrators, causing outages and waste of man-hours.3 4 5
- All of the low quality content that is generated by these models are contributing to the overall enshittification of the internet.
- The marketing message for the non-tech savvy is nothing but snake oil. There’s no actual “artificial intelligence” in sight, yet they easily sell the image of an all-knowing AI thanks to mankind's natural tendency of anthropomorphism.
This state-of-affairs combined with the actual usefulness of the tools that are available, the stand I take is not to use them and to steer away from all the products that try to embed these tools into their ecosystems.
Now comes the prediction:
Do you remember the time when blockchain sounded interesting? Fast forward some years and at some point blockchain hype became so bad that a product embedding a blockchain facet just for the sake of it became the largest red flag imaginable by the consumers.
I see the exact same future for today’s generative models. Not long from now, rolling your eyes with disgust when you hear the word “AI” will be the mainstream reaction rather than a fringe one. We’ll look back on all the generated content and code with a great deal of vicarious embarrassment also known as Fremdschämen. In hind sight, the lengths that people want to integrate these tools directly into their terminals, editors, git repositories and devices will just plainly look dumb, especially when we learn more about the hidden costs of the state-of-affairs I laid out earlier.
That’s my prediction. It might be correct, it might be awfully wrong. The point of this thought exercise is just to “time capsule” my thoughts and see how it ages.