Post #2160482
2026-02-07 10:14 UTC
@pojntfx Still 99.9% trained on top of stolen data, an unimaginable waste of resources by treating public infrastructure as "free real estate" (without respect for robots.txt, by visiting often the same links ten thousand times or using sketchy residential proxies to hamper services like Codeberg's and circumvent explicit "no"'s), and vibes towards the law (which really doesn't apply to the commoners™ in many jurisdictions).
Replies (1)
-
@pojntfx@mastodon.social 2026-02-07 10:17
@n0toose That's not the case for a lot of those open models anymore. Apertus is a great example of it, it was built only from data sources that explicitly allow crawling https://www.swiss-ai.org/apertus "Stolen data" is ofc also very debatable. Tightening copyright law, patents and so on even further to "stop" LLMs from being trained will probably only have negative consequences on those writing OSS. I really don't want to live in a world where you're legally prohibited from learning.