I've been watching the LLM + RAG1 space for a while now. A flourishing myriad of projects a comes up almost every week. One of them communicates particularly well on what is the immediate potential of adding a LLM + RAG1 system on your website.
Best landing page I've seen in a while. You can also find several demos. They achieved a very cool milestone back in May 2023. Adding the LLM + RAG feature to your website is as simple to integrate as a widget.
This feels right, compared to ChatGPT plugins. Think about it — do you really want to go to ChatGPT to book and compare hotels? I don't.
What I want instead, is for the websites that I already know and love to become much better at understanding what I am looking for as a user.
But my favorite websites can hardly compete with OpenAI. They most likely won't develop in-house solutions anytime soon.
These should be the next natural steps for OpenAI / Google / Meta / Anthropic and competitors:
- Work hard to cut down inference costs so that the average cost per user becomes significantly lower that the ARPU2 of most major websites
- This unlocks one big thing: you can offer a large language model widget to any website that wants one, virtually for free
- Make the integration easy-peasy, as simple as adding a Facebook Like button to your website + sharing API specs
- Boom, you're everywhere
A few months ago, when ChatGPT plugins feature was released, some commenters said that OpenAI was aiming to become the App Store of LLM apps. It appears now that it will most likely never happen.
What I predict instead, is that the next big winner in the LLM space will be the one who will create something that a lot of people will want to add on their website for free: the next magic button.
NB: I know that a couple of open-source solutions must be out there already and there may be a chance for one of them to steal the show. Are you building one? Come say hi. I would be happy to feature it on llm.dev.recipes.