39
is K2 horizons historically significant as the first actual open-source LLM, or am I falling for hype? (lemmy.zip)
submitted 1 week ago by to c/selfhosted@lemmy.world
20 comments fedilink hide all child comments
 

i have been following Chinese models for about two years now because they are open-weight and qwen is fun to run on my kubernetes cluster, the news about this Apache licensed model complete with a recipe to make it again with potentially different ingredients is making we want to abandon them for something more ideologically sound and way more interesting

this new model, couldn't universities rebuild it with different contexts and study it in ways you can't reproduce in other models? like, is k2 horizons a good scientific foundation on which to study machine learning?

or am i just lacking way too much context and falling for hype?

you are viewing a single comment's thread
view the rest of the comments
[–] 5 points 1 week ago (2 children)

My question would be if it uses stuff like CommonCrawl as training material. Given their size I assume they used anything incl. unethical training data, but at least admit it?

The only models I ever found that even tried to only resort to ethically obtained data, being FOSS etc. were the tiny ones from PleIAs. And as expected they're completely useless.

So far I concluded that an "ideologically sound" LLM is impossible due to lack of training data. Unless your ideology allows to steal stuff.

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

    I’m assuming a lot of its training data is synthetic and distilled from Chinese models that were themselves trained from pirated data and distilled from American models trained on pirated data. It would be quite remarkable if the training data involved no piracy whatsoever. Then again, it’s open source, so I suppose it would be essentially reversing a reverse Robin Hood.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •
  • [–] 3 points 1 week ago

    I mean, they "plan" to publish that info, right?

    I'd bet they used at least CommonCrawl (with it being the majority of data), arXiv, Wikipedia and the set containing all of Github. Probably not a lot of distillation.

    CommonCrawl is one of the reasons small websites and social instances get DDoS'ed by rules-ignoring AI crawlers. Wikipedia data basically always gets used without paying them. Github… well, it's a prime example of how they broke millions of licenses.

    There's also other stuff commonly used. Don't get me started on the training sets for image generation. It's beyond disgusting (and I'm not even talking just about theft and cultural destruction at this point).

    This stuff is a bottomless pit, and if those (F)OSS models really aim to be up at the top they'll have to break every possible moral, ethical and legal rule just like everyone else. Even more so if they omit closed-source training sets.

  • permalink
  • fedilink
  • source
  • parent