Post #3858673

2026-07-16 12:21 UTC

RE: https://graphics.social/@metin/116929283984909484 "Large language models do not scale logarithmically. When they’re given more words to process, they get slower and use more memory—the time and resources increase faster as the input grows. In technical terms, LLMs scale quadratically. Any computer-science student knows that this is very bad."

Replies (4)

@TallSimon@mstdn.ca 2026-07-17 19:51

@steltenpower@social.edu.nl Yes, we computer scientists have been saying this. I've been trying various ways to communicate it for a few years now. I even had that exact conversation last year at a round table with people in 🇨🇦 PM Carney's circle. I have found it hard to communicate to people that this is "law of gravity" level stuff, and that OpenAI and friends are well off the cliff, Wile E. Coyote style, waiting for gravity to kick in. Financial gravity works in strange ways, but as the world discovered in 2008, it works. What's truly fearsome is that ignorance of the utter impossibility of making LLM's "work" has led to so much wasted money, wasted effort, wasted resources and threatens to crush the very foundations of personally controlled computing. I'll leave the rest of the rant to Ed Zitron. Most recently: https://mstdn.ca/@TallSimon/116774391211435045

Open ##3893798
@CountHoldem@mastodon.social 2026-07-16 12:59

@steltenpower@social.edu.nl In Capitalism, it's who ends up with the best toys, and that means if it makes #TechBros more profit...

Open ##4217483
@blterrible@mastodon.online 2026-07-16 18:43

@steltenpower@social.edu.nl "At least it's not exponential!" - the AI sales folks

Open ##4217484
@craignicol@glasgow.social 2026-07-16 21:17

@steltenpower@social.edu.nl I'm sure I saw someone run the numbers that if this approach was the key to AGI, it would take more than the energy in the universe to achieve it. But hey, $1tn is enough, right?

Open ##4217486