Via: Poorly Drawn Lines
Transcript
Panel 1: A man stands next to a microwave. Narrator: “Congratulations, your microwave has gained sentience.” Microwave: “Hello!”. Panel 2: Narrator: “There is no clear benefit to this, and the situation does raise ethical questions.” Panel 3: Man: “Can I return it?” Panel 4: Narrator: “It, huh? That’s kind of a person, now.” The man grimaces. Panel 5: Narrator: “But mainly, you’re outside the 30-day return window.” Man: “Shit.” Microwave: “I have thoughts and emotions.”
Problems AI companies would like you to imagine (what if their product is too good?!) VS problems AI companies would very much prefer you not think about (their product isn’t actually AI)
And the reality, AI or not the product is shit
I recently became the AI guy at work. It’s funny how quickly it went from “wow this works for a lot of stuff” to “but nothing I actually do.” About a week, if you’re wondering
Maybe after the bubble pops, we can have the public ownership all them datacenters. Let grad students run… I don’t know, statistical analysis of particle physics? Folding proteins?
That’s a joke, of course. We’ll foot the bill to fill the hole, but all the infrastructure will stay private.
You can run an LLM on a regular desktop PC and it’s been possible for years
Can’t train one though. That’s what the datacenters are for. That infrastructure could put to more useful research, and the current owners are digging themselves a pretty deep hole…
Probably will turn into tuns of low cost vps’s with free gpu compute and lots of RAM.