My background:
I work a little bit with JavaScript basically on just doing a few personal websites.
I am interested in using OpenAI… but how can I start working on it?.. and could I use it for what I want?
For example… I want to offer answers about my self and companies and/or products that I work with… Can I use OpenAI to provide information about specific questions? If so… how would that work??
I normally work with Next.js … if I can add a chat functionality to an existing site that would be good, if not I can create a separate one… ALSO… how would the chat work on Progressive Web App site? would it be loaded to the phone?
And yes… interested in how to work with JavaScript instead of Python…
You sure can use JS with the OpenAI API, you might like to take a look at the documentation here
Python and node.js examples are typically available for API functions and there is a great community of js developers here who can help you out with queries, take a look at the documentation and come back with any questions you might have.
Is there a way to teach ChatGPT how to respond to a prompt?
Who are you?
What do you do?
How old are you?
^^ All of these are example of specific questions. Is there a way to teach the bot how to respond to specific questions about an specific person or company?
Sure, to get you started there is a series of short courses from Andy Ng and members of OpenAI staff, these are using python for the API but the methodology remains the same across platforms and will get you up to speed on prompt engineering for software integration.
Also I’ll prime you now that your use case sounds like it would need to take advantage of “Embeddings” these are stored blocks of text that you can find “similar” text to, it’s not as simple as a text search, it’s a semantic search which can take nuance into account.
It is important to clarify that ChatGPT is the chatbot product at chat.openai.com. It already knows what it is, and will be resistant to taking on other roles or names, instructions which will be soon forgotten part of conversation anyway.
With the API, you can write your own system programming message that remains, describing how the AI will operate. The AI you get still has been tuned to give robot answers with AI disclaimers that must be overcome by your instructions.
I was for example asking those questions but not related to the OpenAI chat bot itself, but the chatbot answering those questions as if he was say an assistant to a company or person… or give information about an specific company or person…
I understand you can train the bot with embeddings??? and I think you might need a separate vector database to connect to OpenAI and teach the bot the answers you want for specific questions? like… what is this non famous 3rd person age and what does he do for a living??? Or what products is this un-known company selling?
I am just reading about all of this today …
Also… what would be a good vector database to use for OpenAI??? and yes… I want to do this in JavaScript…
Yes, the AI can certainly be BoeingBot, and answer questions only about selling you airplane parts.
An example within ChatGPT to answer your questions, using a well-programmed personality (that loses its quality since it is only programmed by the user, not part of the system)
The trick then is to not let your BoeingBot be turned into a fairy princess by a language adversary such as me.
A vector database using embeddings is not used to “train” the bot. Instead, it can be a knowledgebase retrieval system that gives the AI some injected knowledge similar to the user’s question, so that the AI is better informed when it tries to answer that question.
