/vt/ - Virtual Youtubers

>>41712432
>Suggestions must be on the brilliant line of thought of forcing companies to audit these models to make sure they are "safe"
models cannot be audited. They can be probed for safety metrics via adversarial training with labeled unsafe content as a reference.
It's less likely that the government will crack down, or look down and chase after rogue users or unsafe large machine learned models. It's more likely that big models, the ones they use in super computers, will be carefully and constantly maintained to stop your BABIES from being told to eat batteries or something
meanwhile, if you want nsfw/funny you just get a (slightly) weaker/faster model and run it on a google cloud server (already very easy with KoboldAI for like $1 per compute hour on 8 gpus)