The most terrifying aspect of this is how these models are might be trained. If the source material is very effectively crafted, you might imagine an army of ag

The most terrifying aspect of this is how these models are might be trained. If the source material is very effectively crafted, you might imagine an army of agents with cult leader like eloquence indoctrinating the weak of mind. Combine this with politically subversive motivations and you have serious threat to societies where the governments do not exert control of people with force. In the past, it was difficult to disseminate disinformation since it was often planted in newspapers and such in other nations rather in the US. At some point this information might make it's way into our information ecosystem years later (false narrative that the US created HIV for example). The internet and social media has already changed that. Now disinformation flows instantaneously across borders, even finding unethical organizations which eagerly acts as clearing houses for this toxic content. We still have not successfully adapted to this crisis. The reality is we have less than a few years to excise the already existing disinformation laundering networks, reinforce our information ecosystems, and educate the public. All this needs to be accomplished before addressing the imminent threat that malign language models pose in the immediate future. This is not only threat from abroad by hostile states but also from powerful self-serving organizations within our nations which might supplant democracy with a domestic brand of authoritarianism. With people like Peter Thiel and Elon Musk bankrolling OpenAI, it's not looking good.