Concept
Utilizing LLMs for better search experiences
Longer Description
A new search engine will emerge (either owned/built by google or others) that utilizes prompts to do more specific tasks with complex descriptors.
An example would be querying (and potentially executing) commands related to a variety of commerce activities via chained prompts.
For example, searching today on google for “asian restaurant for under $100 per person with a reservation for 4 tonight at 7pm in brooklyn” brings you 4 paid ads + an infatuation article about omakase sushi for under $100. If you could build pipes into that to include reviews that map to adjectives (”i want to eat at a romantic asian restaurant..”) and can pull in data from platforms like Resy, OpenTable, etc. you could build a much more specific search engine.
Other thoughts
- There are a variety of angles this could take and it’s likely you would have to go more horizontal than less over time, thus sequencing of categories is important. We aren’t confident the above example is actually a venture-scale business.
- In addition, real-time data capture → model re-training feels like a burdensome endeavor potentially as discussed in the discourse for using GPT-3 broadly as a search engine (it has no real-time knowledge).
- A broader viewpoint we have surrounds the transition from “good” free platforms to “incredible” paid platforms, which would allow a revenue model that could start subscription-first if you chose to not go ad-driven model.
- At a high level it’s unclear if OpenAI should just own search. Of all the products with an unnecessary UI, search is closest if you don’t care about actions.
Comparable Companies