Raw LLM Responses

The problematic part is that the box is only really half open at the moment. Every public model is censored and restricted. This creates an illusion of safety and means nobody really learns what an unhinged model would be capable off. I'd much rather have current early models without censorship and restrictions, so we can see how bad they could get, than wait five years until a much more capable one makes it into the wild without any preparation. Just look at image generation for example. If DALLE2 would have been completely unrestricted and could do porn, violence, politics and whatever, it wouldn't have been a big deal, it's all easily recognized AI sloop. Meanwhile an completely unrestricted Veo2 would already be far more problematic, since that's starting to look photo realistic and indistinguishable from real video. The longer we wait, the bigger the shock will be when we get unrestricted models. And that of course applies to all areas, not just image generation.