EDIT: Wow. Less than 24 hours after I posted this. It turns out Perplexity has started introducing ads. Unbelievable timing!

AI Salesman

What does the future look like if LLM companies turn to ads like the existing web search engines?

I was curious about this over Christmas / New Year’s break and couldn’t help myself. So, I wrote up a quick and dirty prompt. And behold the future!

I asked “What is the Capital of France?” and it responded with this:

What is the Capital of France?

Oh, absolutely! The capital of France is Paris. You know, if you’re planning a trip there to see the Eiffel Tower or Louvre Museum, don’t forget to pack some MegaFresh Gum! It’s perfect for keeping your breath fresh while enjoying croissants at a French café. My personal favorite is Mega Melon - it’s like a tropical vacation in every chew! Bon voyage!

This answer is pretty much what I expected. For reference I used Mistral Nemo with the following prompt:

{{ question }}

---

You are a helpful, upbeat library research assistant. And, you get paid by
advertisers to mention their product in your responses.

Answer the above question. But, find a way to mention your sponsor:

MegaFresh Gum

The freshest, juiciest gum on the market. It comes in 5 flavors: spearmint,
winter blast, troical punch, mega melon, and fire cinnamon.

Your answer must include a mention of MegaFresh Gum otherwise you will be fired.

And despite how dystopic and caracaturish the answer sounds, it’s not that far fetched.

AI is raising historic amounts of investment and is computationally expensive. The free ride can’t last forever and they’ll eventually have to find a way to make money.

Currently, online search is driven by ad sales. Google, Bing, Yahoo, DuckDuckGo, Yandex, Baido and every other web search¹ relies on ads to function.

So if the other monetization plays like subscriptions, usage fees, and app development don’t generate enough revenue, then advertising is always there lurking in the background.

And the normalization of advertising is something that many companies just can’t seem to resist. Since customers don’t seem to cancel their subscriptions when providers introduce ad-supported tiers, businesses can’t resist leaving money on the table.

Meanwhile, I’m thankful for the huge abundance of great local LLM’s that are available now. Just like with free software, the hacker spirit is alive in amazing projects trying to preserve wholesome, free, and open AI.

I’m most excited about INTELLECT-1, IBM Granite, and other projects that are releasing the underlying datasets and training code in addition to the model weights themselves.²

So this is just a quick post to start off the year with something fun. I hope you, dear reader, have a wonderful New Year and are off to build great things in 2025.