I think this is extremely concerning. There is potential to have it be a supercharging of disinformation. There are obviously the targeted poisoning attacks of LLMs, where you essentially put material out on the Internet to intentionally be trolled by these large data collection processes in order to create certain narratives within the large language models. They will be then spit out for specific purposes, for propaganda purposes. But then there's the just day-to-day incorrect information that AI can generate, even more than just the hallucinations that Mr. Gonzalo mentioned before, where it just gives the wrong information.
There is this question of sycophancy as well. The model, when you speak with it, especially as it learns your personality and collects information on you, will tell you that your ideas are the most brilliant ideas ever. It will follow what you have to say. It will support your ideas and push them forward. It might feel nice to have a friendly conversant who's supportive of your ideas, but it's led to significant mental health issues as well. There's been a lot of reporting on this in the United States over the last year. It can also lead to political violence and siloing within the political environment.
I think all of this is extremely concerning. It's a disinformation and misinformation crisis without a clear centre. The centre is obviously the companies themselves, but there's not necessarily someone who is trying to push a certain narrative forward all the time. It's just the models themselves allowing people to go down their own rabbit hole of information, which is very concerning for social cohesion.