I see that the market strength of RAG-based solutions is in internal (proprietary) repositories of data.<p>I built a small one for publicly accessible documents but the models (or models with internet access) are gaining ground.<p>I wanted to hear more thoughts.
RAG is going to get eaten alive by Google, MS, and OpenAI.<p>The only way you will be able to beat them is in tiny niches, but when it comes to importing an organizations "knowledge" it will be hard to compete with the companies that already have it (O365, Google)
For all those saying "RAG is a feature" I mean search is a "feature" of a bazillion systems, with Google, etc offering services to search over your data. And a thriving ecosystem of open source and commercial vendors. Search itself has vastly different user needs from job search, to ecommerce, to enterprise. PDFs aren't products aren't jobs and they all have very diverse use cases.<p>BUT - I do think its way way way overcrowded vendor market right now. The vector DB vendors are all RAG RAG RAG and have payed little attention to the fact they are actually a search solution focused on one type of index, and there are a million other applications of a search engine.<p>So its a crowded dumb market, but it probably has some kind of future.
OpenAI already offers a limited RAG solution, and it's only a matter of time before they're a product.<p>I would have paid a good sum for it a year ago, but we figured it out. I think it's like setting up databases. There was a market for that kind of consulting 15 years ago, but in the long run it might not be there.
In the market of AI solutions, RAG is a feature. The best way (imho) for small outfits is to have an existing expertise in a field and bring these tools to it. The big players will capture the generic, one-size-fits-all market. Specialized solutions for specific problems and industries is where it's at. Also, don't limit yourself to LLMs and RAG. There is a lot of other algorithms that are useful and provide value