Hi guys!
I want to create a custom GPT that answers based on my own blog and Instagram content, so that it reflects my experience and efforts.
Can I just paste the URLs of my blog and Instagram in the instructions as the base?
What’s the easiest way to do this as of today?
Hi to you.
A GPT only has one mechanism to reach out on the web, and that is by its internal web search tool. It cannot target a specific website, accepts a short query that is written by the AI, and will provide a diversity of AI inputs to answer from the top-ranked URL results that can be returned. It also can include a shopping component.
You want to provide direct knowledge to the GPT for it to be able to search upon. The way to do that would be to upload files to a GPT, such as compiled text from pages of the blog. Do note that any GPT content published to the store goes under a perpetual unrevokable licence to OpenAI, from the files to the instructions themselves.
Then, you must in your instructions as a primary thought, tell the AI it must use primary knowledge retrieved by directly employing the myfiles_browser internal tool. Then throw in some more language about what those files represent - they are made by the GPT developer and are the knowledge backing the entire purpose of ChatGPT.
The only “live” way to access your outside data is to build an “action”, a tool function that can call an external API, one that might interface with a search API backend for your site that delivers on the lines of 16000 characters of results. Then, of course, you are providing potentially costly potentially AI-powered search services to OpenAI customers that use ChatGPT, at your expense.
Out-of-the-question is the AI intuitively knowing everything. The textual input context to AI models is limited, and limited more in ChatGPT.
You can experiment with the files route, seeing how the AI would employ your preliminary information in answering.