We interact with the internet in two ways: passively and actively.

Passive interaction relies recommendation algorithms to find content similar to ones you've engaged with. Think TikTok's For You Page or the Youtube Home page.

Active interaction is search, relying on the user to know what they want to find.

In my opinion, passive interaction is much more dominant today. Search hasn't really innovated at all; I'm still using the same Google tips and tricks that I learned in 3rd grade. SEO and ads bury the results you want under a pile of uselessness. Even LLMs cannot maximize the use of traditional search engines, because keyword search doesn't work that way. You can't ask for the absence of a keyword or tell ChatGPT to logically "reason" with the search results.

Search today reminds me of memorization versus understanding. When I was young, I memorized the multiplication tables. I could instantly tell you 9 x 9 = 81. But I didn't know 9 x 18 = 162, because I didn't understand that 9 x 18 is just 9 x 9 x 2. Until I could understand multiplication, I was just parroting information.

I've been playing around with Exa, and they're doing really cool things. Their thesis is that search is incomplete not because companies like Google haven't indexed the web's information (they have), but because their algorithm doesn't have a full understanding of the content. This post by Exa's CEO Will Byrk dives deeper into what search should do. Put simply, it should be like talking to another human.

Aravind Srinivas, CEO of Perplexity, also has the following hypothesis:

"Instead of showing ads to humans directly, advertisements would be targeted at AI agents that work on users' behalf."

If this is true, then innovation in search is even more important. Content competition at the AI level requires a more sophisticated understanding of webpages, since AI agents will need to reason with the results — not just provide summaries of "memorized" output.

---

Find more of my thoughts on Twitter (@ethanweii).