Worse yet is when AI gives answer that are ads without knowing it.
Not long ago I asked ChatGPT for the best washing machines (or something). It gave me a list with a little information about each one. I then asked for its sources. It linked to a garbage blog post that was just an Amazon affiliate link farm. There was no research, no testing, nothing... just random links to try and generate a few cents per click. This is the "knowledge" we often get from AI, without knowing it.
I once had chatGPT run a research about popular stacks in job openings across Europe. Not that I don't already work with React + some Python, I was just doing it out of curiosity for it's results.
After 5-7 minutes of work, it returns many results, yet it's citing 2 specific websites as sources, one of which was blogspam you'd write to get visibility on Google results.
So I guess we're heading towards a future where websites will be optimized to increase the probability of chatGPT and AI tools to use you as a reference and link to you with confidence, regardless of their sources.
I had a recent example of ChatGPT giving me a really fishy answer I didn’t believe after it searched online, so I looked at the cited page, and it was clearly hallucinated slop written by another AI.
I wish for it to only use sources that are older than 2019 and have zero ads and referral links, haha.
That first sentence gave me shivers because I know it's true. I don't think we realize the extent of the subtle but constant manipulation we'll all get to experience.
Subtle manipulation maybe. Subtle ads do not exist. Theoretically it's possible, but I've yet to see one. Advertisement is blatant. Not that it doesn't work (on my as well), but it's blatant.
In the above example, someone is paying for them. "In 2021, the agreements between brand owners and films and television programs were worth more than US$20 billion".
I'm pretty sure you are very unfamiliar with how ads actually work.
The best advertising is word of mouth advertising and smart marketers seek out people of influence in their communities to spread their products. This was well known in marketing long before the term online influencer was a thing. It's very hard for most people to even notice this kind of advertisement is even happening.
My first reaction when a friend tells me about a product is, that he/she must've seen it on social media.
- Do they actually own the product?
- How long have they owned the product?
- Show me how it works.
- How much have you paid? Else it's worthless to me, but I'm happy for him/her
Huh? I don't think there is a person on earth that does not know the BMW James Bond drives, is not sponsored. Is this really the expectation? I don't believe it.
Have you ever watched someone use Google? Most people look at and click on the sponsored links as if they are organic search results.
Product placement, especially without specific calls outs, are something subtle that most people don't notice. Something like the boxes of cereal sitting on the shelf in Seinfeld's kitchen. Are those ads, is it just set design? I don't really know.
There is also car choice in a movie or TV show. The studio isn't going to design and build an actual car just to avoid using a company's product. Which car do they pick and what does that communicate about the brand to the viewer. Is this an ad?
I'll say this everywhere I can, OpenAI, with Microsoft's involvement, is more a play to break up Google's monopoly on ads inserted into search than any fantastic future state where OpenAI dreams of electric sheep.
You could see this in the agents demo. Need a suit. Ah, let's check J Crew. You'd like that, wouldn't you, J. Crew? How much would you pay to make sure it always checks your site first?
Showing tracking-banners ("cookie banner") that hide their "reject all" somewhere in sub-menus of custom settings is also illegal in Germany (and the EU). Yet you see them everywhere.
Why would you need to retrain the model or update the SFT? You could just dynamically update the system prompt to include things it should advertise.
You could even have something like an MCP to which the LLM could pass "topics", and then it would return products/opinions which it should "subtly" integrate into its response.
The MCP could even be system-level/"invisible" (e.g. the user doesn't see the tool use for the ad server in the web UI for ChatGPT/Claude/Gemini.)
LLM's have already digested all of the web. There isn't much new data for them to consume. It is rapidly moving to synthetic data anyway. The limits of human information have been nearly reached, from a consumption POV
And without the web there is no new datasets for AI so it’ll grind to a halt.