21
Ramblings from a non-practitioning AI hater (blog.nocturnal.garden)
submitted 2 weeks ago by to c/anarchism@lemmy.dbzer0.com
7 comments fedilink hide all child comments
 

I wrote this, and while Anarchism isn't directly mentioned, it's a foundation of my values. I know this is kinda an AI instance but I still wonder how y'all treat the discrepancy between the externalized costs and using AI for whatever.

you are viewing a single comment's thread
view the rest of the comments
[–] [S] 7 points 2 weeks ago (5 children)

We do not promote corporate (e.g. OpenAI, Midjourney etc) and datacenter-based GenAI as we recognize the harmful effects their creation and inference has on societies and we reject fascist ideologies underpinning them.

I thought at least the creation of the Open Source/Weight models has the same issues? I'm still looking into local options but don't they also need to do some very heavy scraping?

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 5 child comments
    •
  • [–] 5 points 2 weeks ago (4 children)

    The point is that our use of those models does not provide any return of the investment on the costs incurred by training these models. We can't help with those externalities. By the time the models are released to us those are already done. We of course do not support these actions.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 4 child comments
    •
  • [–] [S] 5 points 2 weeks ago (3 children)

    So you'd prefer they don't create new ones, but you're using those that are built already? I get that, I'm just not sure how much the creation of new OS models is driven by how much the current ones are used.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 3 child comments
    •
  • [–] 5 points 2 weeks ago (2 children)

    I'd prefer if they create new ones without aggressive and asocial scraping.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] [S] 4 points 2 weeks ago (1 child)

    I didn't know that's possible, though they can of course make the scrapers a bit smarter than scraping every possible link a website offers. I wonder how much of a difference that makes

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •
  • [–] 4 points 2 weeks ago

    There is a lot of really low hanging fruit.

    Right now there are a lot of aggressiv scrapers, sharing little to non of the work they do. A single aggressiv scraper shared by many or all model-trainers would probably never have been a big problem and would have had the same output.

    We also already have a public, scraped archive of the internet, The Internet Archive, who already did the annoying scraping part and already have a programmatic interface to that (absolutely fucking huge) database.

    My point is, think of these companies like all other companies that make money from extraction of ressources. They are not doing "a rational minimum" of sustainability, not even to keep making money, they just pump everything dry and move on.
    A bulldozer can drive around every tree in a forest without doing damage, but a bulldozer in a forest is probably there to kill the trees.

  • permalink
  • fedilink
  • source
  • parent