Freya HolmérJan 4, 2025
Creideiki
As a technical person, I usually critique generative AI from a technical perspective - its use of unsustainable amounts of electricity, its lack of world model meaning its output is never going to be correct, the business models not adding up, training material getting harder to find now that much text on the Internet is AI slop, how its purported connections to AGI are impossible sci-fi smoke screens, etc.

However, we forget the humanities at our own peril. I found this latest video by @acegikmo to be worth watching for a critique from a social and artistic perspective.

And also for a reminder of how much barely coherent AI slop is out there when you actually count it, instead of almost subconsciously discarding it out of hand. Finding information in a field you have no prior knowledge of must be a horrifying experience nowadays.

RE: https://mastodon.social/@acegikmo/113763950485888985
Jan 4, 2025 at 5:11pmWeb
Show threadpandoraJan 5, 2025
@creideiki @acegikmo google search became really useless within the last two years. It’s like searching through a pile of trash now
Show threadNiceMicroJan 5, 2025

@creideiki I did not end up watching the whole video, but it seemed to me that the first 15 minutes or so is mostly showing websites which had the kind that existed well before the current LLM explosion.

The fact that everyone wanted to play Google's rankings, lead to the fact that Google is polluted with useless crap that is just simply SEO farming, I guess LLM generated content makes the problem much worse, but it's not like this came out of nothing.

Show threadJari Komppa 🇫🇮 (has moved)Jan 5, 2025
@creideiki @acegikmo thing is, these garbage pages existed before llm, but now they're more difficult to filter out, and Google isn't even trying anymore.
Show threadFrank Hightower 🐁Jan 5, 2025
@creideiki @acegikmo you might want to be careful, a tech bro could read this as "we just need to pump money int world modeling now!"
Show threadCreideikiJan 5, 2025
@FrankHghTwr Luckily, world modelling is something deep neural networks are absolutely crap at, so any progress would depend on academics sitting down to quietly think about things. As this will not result in something flashy to wave in front of the credulous world press within the budget quarter, I don't think the current crop of grifters will care to even acknowledge that the problem exists.