In my last post, I explored the possible bull case for ChatGPT eclipsing Google’s dominance. But I tipped my hat also at what I think is the bear case. As I mentioned, ChatGPT’s unbounded interface a bit like as if, instead of DoorDash’s focused launched strategy, DoorDash launched across all US cities and categories of goods, all at the same time, and when you order something, does its best to deliver it. It shouldn’t be surprising then that while ChatGPT can feel like magic for some results, it gets a lot of queries wrong. You can see what’s possible and seemingly eventual, but the interface doesn’t guide the user to what’s best now.
This to me points to a tremendous opportunity for startups using a more focused approach (albeit still dependent on OpenAI and other’s APIs).
I’d guess these companies will have two attributes:
More focused interface that guides users to leverage Large Language Model’s (LLM) strengths.
To me, this is areas where either an answer being plausible is sufficient, or the answer needs to be precise but there is a measurable correctness of the answer which would enable a closed loop feedback system.
Something like this:
You could certainly imagine a StackOverflow-type company, a disruption to LegalZoom, incredible opportunities in medicine, a new approach to Wikipedia and the news, etc.
Something like a top ten list on the other hand seem problematic to me (sorry BuzzFeed!). There is just too much subjectivity in the answer without more granular controls.
Closed loop feedback system.
One of the incredible strengths Google had with its interface is the closed feedback system. Type any “top 10” query into Google, and while you might get a zero-click answer at the top of the search results (Google’s estimation on the best answer), the nature of a search interface is that Google provides a possibility of answers. For this category of query, each result often has a human author with their own biases and criteria, and it’s then up to the user to find the answer they feel is best. The benefit of owning the search interface (and often browser) is that Google can then analyze this user engagement clickstream data (conversion rate, bounce rates, time away, etc.) to assess and therefore tune the quality of the results over time.
ChatGPT’s feedback loop is not as strong. It provides one definitive answer, and yes a user can give a thumbs up or down, tweak their prompt, ask ChatGPT to recalculate the answer, or give explicit feedback back, but this is a lot of work relative to the implicit feedback loop Google had, which means ChatGPT ends up getting a lot less data for the same number of users.
If a startup is able to design an interface that has a stronger feedback loop thanks to its focus, I’d imagine that would be a compelling network effect over time.
As always, the nature of these weekly posts is that the ideas here are still raw and forming. Curious what you think (and as always, who I should be meeting!).