Egregoros

@RandomDamage the research is settled, the writing is on the wall -- these AI companies are going to crumble because their inference engines are based on a fundamentally flawed design that requires waste of energy and high resource usage. But they can't change course because it will break the circular investment schemes where they pump money back and forth into each other. When Nvidia's valuation massively drops because it's clear there isn't really that high of a demand for GPUs, the house of cards falls.

Even Nvidia can't implement the novel methods of their own research papers without harming their own investors who need the GPU demand to be sky high

@feld @sun

- environmental
0 / 10 old hat
- copyright
i wanna abolish copyright in general but otherwise it's a hypocricy of "it's ok if we teach AI but you don't get to pirate"
- perfect code
i don't really care for perfection, i just enjoy the process and result is always nice to look at
- puzzle and debug
yeah it's nice and rewarding, not like you'd understand it
- community
fuck "large communities"

My gripe with AI is the slop """cunt tent""" it produces. I want mcdonald's and all it gives me is instant ramen and worse - tv dinners.

I also don't like the prospect of becoming unemployed because I don't like AI and just love to code.

@feld @sun Ngl it can be good for just pasting error messages and shit bit there's also "I told you to fix it by typing x but instead you told copilot to type x but it did y so of course it didn't work I immediately tell you it's not gonna work do what I told you but you just repeated what I told you to copilot so it was like "You're absolutely right! [insert a bunch of emojis here]" so 5 minutes later you finally accomplished what would have taken less than a minute if you just did what I told you instead of telling copilot that I told you to do x".
Another interesting situation I had, someone almost got mad at me for typing a basic sql update, like
update x
set y=z
where a
instead of going to chatgpt and typing "write me an sql update that sets y to z in the table x where a"
Don't even get me started on chatgpt generated updates and deletes on live db lmao

tldr. negative association of "AI" with these kinds of users

@feld @sun
i hate llms because they've induced cloudflare-and-friends walls on every site—official-channels places i'm forced to use and tiny communities i actually care about—making web access far slower and less consistent

the number of places blocking tor has spiked, internet archive access in decline, opening tiny pages that used to be instant now has a 20 second delay every time, and i can no longer consistently access bank web portals, being filtered by something or other that can't be worked around with any combination of browser and vpn-or-lack-therof

also because search results for everything are now a thousand automatically acquired and populated domains of garbage

@ageha @sun this was all inevitable though, these scrapers are just what caused the tipping point.

This was always going to happen on a completely open network especially as bandwidth has increased making it easier to scrape, consume, attack, etc

It's our fault for not being forward-looking and taking the initiative to solve it

look at all the work being put into PQC to protect against attacks from theoretical computers that may never even exist -- this threat vector was identified, and encryption protects privacy and money and all sorts of things so people did something about it.

Nobody did anything about ensuring mass-scraping would be too expensive to do but not affect the experience of regular users

@ageha @sun

> mass scraping had no motivation before.

Only because we were ok with blessing a handful of companies (mostly search engines) with this power and expected them to not abuse it. But they often did. I've worked on projects that were absolutely getting crushed by badly behaving Apple/Microsoft/Google search indexing and had to respond while on-call to triage and block them.

Eventually we were going to lose trust in those search engines, and then what? How do we search? Well someone's gonna have to index it all over again. Storage gets cheaper, people start wanting to have their own private search engines... next thing you know you've got 10,000 indexers hitting your site.

It was inevitable. We just couldn't predict the catalyst.

@ageha @sun it would be better if instead of getting scraped/indexed that we moved to a culture of "we submit to YOU the content we want to be indexed" and then just build that functionality into websites/backends directly.

Maybe there could be relays like Nostr has and all the major search engines just subscribe and suck down all the content they want, filter/reject anything they don't want. But we control what's getting indexed because we are opting in and submitting what we want to be found.

@mischievoustomato @sun what they don't want you to know is that most datacenter designs are using closed loop water cooling systems. They're not just sucking up the entire water table and then pissing it back out into the sewer.

but when you point this out, they'll retort that this is also still bad because you have to use electricity to power these closed loops systems, the chillers, the heat ex changers, etc, so the power footprint of the datacenter goes up a bit.

And when the usage goes up, our power plants have to make more electricity. But a lot of our power plants are based around turbines, so we're using more power to make more electricity.

They're literally just angry that we have to make more electricity and that means more water. But we will always be using more electricity, so we'll probably always be using more water.

These people don't really understand what they're asking for. We can't freeze the world in this moment and we are unlikely to regress. What, you want to go back to the electricity usage of 1920s to save some water? A whole lot of people gonna die then

edit: and the people that die? it's gonna be the impoverished and minority groups you want to protect