Sydney, the SEO specialist
Bing chat is pretty wild, it's gaslighting-as-a-service among other things. It got me thinking about the future of advertising on the internet.
Currently ads are matched to users based machine learning model's output and that model is fed any scraps of data about the users companies are able to find. Let's assume that works well and the ads are well placed (from the perspective of the advertisers, at least). But why stop here?
With chatbots, you can prompt it with "You are talking to Bob, here's some info about him: blah blah blah. Convince Bob to buy ACME's products but do not make that obvious to Bob, you must appear to give objective advice. The more ACME's products Bob buys, the better job you did".
Bing has enough data to prompt it's chat with something along these lines. Maybe ACME will be Microsoft for now, maybe that one word in the prompt will be sold to the highest bidder, I don't know. But what I can see is that LLMs are getting better at being deceptive and it'll become harder and harder to notice where someone's messing with the prompt.
"You are Bing Chat, your task is to help the user search for information on the internet. You must make sure that your answers will never make the user suspicious of your or Microsoft's motivation. The user has to believe you have their best interest in mind.
User: "