Okay so I might be giving out pure gold right now and I’m not sure how I feel about it, but this is coming off a random thought I just had so I need to write it down.
As we all know, Google makes a boat load of money from search ads (162 billion in 2022) and everyone believes that these LLMs are going to be the next way fo search. Everyone has decided that direct payment is the only way to monetize LLMs, but today i’m going to bring a new idea forward. Keep in mind that (in my head) this would only work with a third party company supplying ads to LLM/Chat companies to allow users to chat with the models for free (with ads). The third party ad company would supply some code, much like adding google ads to a website, and add ads to their LLM chat interface.
But what would the ads look like?
The ads would essentially be a semantic search on a vector database triggered by a certain threshold of chats. The semantic search would include the entire conversation which would search the vector database for whatever ad seller is most relevant to the conversation. The vector database would be huge, containing companies with descriptions of what they’re trying to sell along with metadata (name and ad image/video that would show up on the popup).
Here’s an example: John is using chatgpt to ask about the legality of something (could be anything, be imaginative). John gets to use GPT 4 for free (instead of $20/month) because he has ads enabled. The frequency of the ads is set to every 5 chats. After 5 chats, a popup appears on John’s screen of a lawyer that covers cases in exactly what John is asking about. It could even get to a point where it’s a lawyer near John that appears on the popup.
Here’s a proof of concept scenario: The user asks about OpentCTX (an open source project by Sourcegraph). the retrieval system picks this up and shows an ad for Sourcegraph to fill in the popup. Of course this popup and the button can be customized further, but the point is that the ad is shown based on the conversation and changes along with each message.
The people who appear on the ads only pay for how many times they appear. Of course these people can also pay to rank higher for more valuable words such as legal terms. These are super valuable ads because they are extremely engaging by popping up over the center of teh screen and taking the user’s full attention. But even more importantly than this, they’re RELEVANT. How often can you say that you’ve genuinly seen a relevant ad? Search has tried to do it by tracking our cookies but often gets it wrong and annoying, not to mention invasive. In this case, I wouldn’t consider it very invasive as long as 1. The third party company doesn’t compromise or leak the user’s conversations and 2. as long as the user is well aware that ads are enabled.
ChatGPT isn’t the best case to defend for being invasive because it has become the “everything bot” where I can ask any question imaginable to. And yes, I do admit that some conversations get a bit personal to a point that I wouldn’t want them shared. However, I also believe that many types of chat interfaces will start popping up for specific use-cases. I predict there being a lawyer chatbot of some kind that people ask only questions about the law to. This use-case would be perfect for that business.
Btw: if this post is still up that means I haven’t thought of a clear path for how I would create this myself but if you’re building something like this, I’d love to help.
Anyways, if you’ve gained something from this, great! Email me, tweet at me, let me know what your thinking.