#llmban — Public Fediverse posts

I see a lot of people complaining about LLMs/AI here and the evidence for harm is overwhelming, but little action. I think an EU citizen's initiative for a ban of AI could be a good start…

So let's think what we need and what the carve-outs should be. It is important to approach this objectively and actually allow for uses that are legitimately very hard to attack.

Once we identify those we can petition for a full on ban. Yes, governments can do that if they choose to. Just treat LLMs like child pornography: If you have it on your device, you are in trouble, if you spread it, you are in trouble, if you use it, guess what…

Make AI-companies whose services can be accessed in the EU into criminal enterprises and prosecute them accordingly.

Carve out pathways to enable access to the very limited exceptions, like certain forms of accessibility-software (AI-generated alt-texts should not be allowed to use on images, but as a blind user you should be able to get a permit to run a purpose-made local model that generates a guess at an alt-text in case there is none) or in some settings where the software might be too useful to fully ignore but too dangerous to make public (akin to software that simulates nuclear effects and can be used to develop nuclear weapons).

Consider providing publicly funded and developed single-purpose software with a license that allows its use, but not its modification for broadly applicable models such as translation software.

The focus would have to be on treating LLMs as an information hazard that must be kept controlled in such a way that it cannot substitute actual thinking or creativity in humans. Only skills for which we are actively fine with them deteriorating in the general public due to a well thought out analysis that reveals that they are obsolete in general must be allowed to be substituted.

After that the point would have to be to collect support and actually campaign. Thoughts anyone?

#noAI #LLMBan #AI #LLM