I’m at OpenAI working on GPTs and would love to learn more about how people are creating and using GPTs for their personal use. If you’re interested in chatting, can you send me a message with some light details on the types of GPTs / your workflows?
I have a custom GPT which I’ve connected with shell access to my home PC with a (heavily) modified version of this old open source ChatGPT plugin: GitHub - etherlegend/gpt-terminal-plugin: Universal command line plugin for ChatGPT
Knowledge files and instructions cover the detailed hardware/software makeup of my system.
This allows me to chat with my computer and the GPT can give very specific answers to questions, rather than broad generalizations or a series of instructions for irrelevant operating systems, etc, without needing to remember to type it all in every time.
The shell access gives the model the ability to do things like restart a docker container, open/close ports, etc from within ChatGPT which is considerably more convenient than other ways to do these things when I’m out and about.
Good to have a conversation with you. I am building GPTs in the pharmaceutical sector that I hope clients will implement themselves. As this is a highly regulated market the instructions and the knowledge files are critical. Most of the time the output is OK but too often instructions are bypassed and information in the knowledge files (usually json) is not ‘found’.
I guess what I’m seeking is that instructions are dealt with more procedurally and knowledge is retrieved more faithfully.
On the positive side I see GPTs as being capable of some amazing things and, when the reliability is improved, envisage substantial usage.
I’m building PoshogGPT to analyse session recordings. I don’t want to spend hours watching session recordings of my users to identify why things don’t work on my product. ChatGPT is very good in analysing these sessions.
PostHog provides API to retrieve session recordings in JSON format and GPT 4 is brilliant in reading that format and provide insights. The only problem is rate limiting of GPT 4, I get blocked very fast as video recordings are heavy. Due to this I’m writing the code to use GPT 3.5 for the same instead of using ChatGPT interface. Happy to talk, if you need more insights.