we have three access levels to AI:
1. community models (huggingface)
2. blackbox models (anthropic, openai)
3. corporate inhouse models, inaccessible to the public
we're mostly angry at (2), but (2)'s knowledge can be sampled through APIs and lands in (1) and (3).
when funding fails, and it will, then we're left with (3). (2) is gone/paywalled and (1) will be scattered and siloed because huggingface is paywalled.
this is bad because AI is rapidly improving.
@lritter one problem with the IP objection to AI (if focused on exclusively) is that Disney probably has enough internal IP by itself to create as much image and video model as anyone could need. So if we kill scraping-based AI, potentially AI survives, but now Disney has a monopoly on it.
This is one of the reasons I try to frame AI as just a special case of the harms of our approach to IP, rather than in opposition to IP.