I’m really glad to hear that Austin has been playing AI Dungeon. I got into it about a month ago through the random out of context snippets being shared and it really is something unique. At the same time however it is horrifying.
The natural language models are progressing at a faster rate then I thought and it’s getting to a point where in probably 15 years you will not be able to tell if it was machine generated or not and it is going to cause a lot of trouble.
Here’s one I just generated using InferKit (that default prompt…), it took maybe 4 rolls before I got back text that I liked but that is something you can train the model on if you are building something specific like say scam/phishing messages.
You know what’s also scary? Sales software. Everyone worries about things like the government knowing too much about them or their ISP (which are things you should worry about) but the private sector has them beat and it calls itself CRM. They’re the kinds of companies that are secretly massive but have maybe 3 paragraphs worth of info on Wikipedia because they would prefer no one knows about them outside of their intended users in sales. Here’s a screenshot from one called ZoomInfo
The amount of data they have on individuals and their lives is just not right, especially for a privately held company that does not have to disclose to you that you exist in their database unless you go asking and can prove you live in a country that actually cares about privacy.
So here’s my question, these companies are already working with the biggest of data and are already trying to devise ways to get their clients all the information possible to sell their targets the next hot cloud tool but what happens when someone starts to use these systems plus a system like AI Dungeon or InferKit to preform phishing and scams at a massive level?
This doesn’t even begin to touch on just misinformation campaigns in general. How do you even begin to tackle that as an issue when your false news can literally be written and fact checked by bots working in unison producing text that looks believable across several sites and social media platforms? The 2016 and 2020 U.S. elections were already bad, everything past that is only going to get worse as the tech gets better.