English
Share
Sign In
Creating a service with a language model
Prompts play a big role in streamlining interaction with GPT models such as GPT-3.5 and GPT-4. I believe that as artificial intelligence develops in the future, it will simplify communication with foundation models and make it easier and more efficient to use them in a variety of applications. As artificial intelligence becomes more common, it reminds me of a time when planners studied API and SDK concepts. It's not necessarily something you need to know, but it's something that's good to know. In fact, I think what people who create or will create artificial intelligence-related services need to know is structure. The most convenient thing is that if something like OpenAI sends an API, you can receive it, input it, and print it out... Typically, representatives or service providers will not prefer this because they want a better experience or cost advantage. Even if it’s just to quickly lead the market. A representative example is the domestic cloud storage business. In the end, if you build your own IDC or data center, the operating cost itself will be reduced by more than 50%, but it is not easy to build and it is not easy to design it properly.
Personally, if anyone is planning or dreaming of an artificial intelligence-related service, regardless of occupation, I would like to read the above information. In fact, it's good to have a clear understanding of just three concepts. These are Foundation Model, Embedding Model, and VectorDB .
Emerging Architectures for LLM Applications from Matt Bornstein and Rajko Radovanovic
The Foundation model is like the librarian in this library. This is a role that knows a lot about books (information) and can help you find the book you want. When you ask “Tell me about computers,” it recommends books about computers.
The Embedding model is the task of converting each book into easily searchable code. It's like putting a unique barcode on each book. This way, you can find the book you want really quickly.
VectorDB can be said to be a large bookshelf that organizes books using barcodes. These shelves keep your books sorted by topic or any way you like. So, if you say, “I want to read a book on this topic,” we can provide related books right away.
When creating an artificial intelligence service, the questions people ask AI can be categorized more easily than expected, and there are clear areas that people ask about frequently. Understanding this allows you to design VectorDB and Embedding models appropriately to significantly reduce costs.
ⓒ 2023. Haebom, all rights reserved.
The source is indicated and may be used for commercial purposes with the permission of the copyright holder.