Possibility of chatGPT

so my requirement is like this

  1. I have 100 paragraphs
  2. I want to put 100 paragraphs to database
  3. when the user asks something, the engine will automatically answer by looking for answers in the 100 paragraphs earlier

example:

  1. I have 100 paragraphs about ‘PHP programming learning materials’
  2. I input all the learning materials into the database
  3. I asked 'how to replace the string`? then the engine will answer by looking for answers in those 100 paragraphs

so my requirement is engine that without training phrases, the answers can be automatically searched in those 100 paragraphs.

is this possible in chatGPT?

ChatGPT, not reliably; GPT-3, absolutely. I believe text-embedding-ada-002 is the way to go. @raymonddavey just informed me about its use cases here:

2 Likes

Embedding will find you the context you need to answer the question.

Then it will depend on how you phrase the final question you send to GPT. You may have to send the request to a CodeX model if it is PHP. But Davinci might be able to help too

I assume you want it to modify the code for your use-case. In this situation, a prompt to Codex would probably be the best bet

1 Like

Thank for your response @raymonddavey @Dent

Since your response, i learned and tried embedding ada and davinci for about 150 paragraph. i tried to ask via prompt and the answer was match as my expectation.
Step that i did is :

  1. i have 150 paragraph
  2. i Convert it to vector (vectorization process)
  3. save vector as csv file
  4. Use csv file to be called by embedding

Is my step right?
If yes, if i have more 1000 paragraph, So i must convert all to vector?

Is posible to use embedding without ‘vectorization process’ ?
because vectorization process need big effort

The purpose of this question is make sure my step is right. I worry when i finish to convert 1000 paragraph but any technique that allow to use embedding without vectorization process

Yes your process is correct. You might be able to combine paragraphs so you have less to do. You could do this if the paragraphs are related

Otherwise you can do what you described and your searches will find the best matches