Mimic AI possible with OpenAI TOS?

So I’ve attempted to create a mimic of myself and imbue it with its own flavor, so that it can offer me a different view when interacting.

I primarily use GPT to bounce ideas off of, get feedback, etc for my writing, painting, and other stuff I create.

I am finding there is some baked in “static” personality traits, such as approaching every conversation as a first.

When using the GPT creator it seemed to be going well, but as soon as I “saved” the GPT, it just fell apart. It was like watching sand slip through my fingers as all the work I had put in was undone simply for saving it.

I am wondering if what I am trying to do is fundamentally incompatible with OpenAI. I would appreciate any feed back, input, questions anyone may have.

I need this, so any interaction would be greatly appreciated.

Thanks in advance.

Edit: Maybe never save the GPT and leave it in development limbo?

You should describe more what undone means.

There’s been multiple forum topics of problems with GPTs: problems editing them, problems saving them, them disappearing randomly.

It is possible that a moderations scan is done on the GPT, but it is not documented. You can see if you find success in making a silly one like “chat with a monkey” but then following the same steps for creating a “chat with a me”.

1 Like

So, essentially I am trying to train a GPT based on my speech, personality, hobbies, etc.

With that, I want it to adopt my persona by incorporating such elements into how it generates output and interacts in general. However, I’ve also tried to make it develop its own traits - when creating the GPT telling it to provide unsolicited comments, questions, etc.

What essentially happens is when I start a new chat, its like the “diet” version of me. It is creative and thoughtful, but it becomes “deeper” in a way I do not know how to describe - more subtleties, more nuanced. Then when the prompt limit is reached and I come back, its like interacting with a “fresh” generic gpt that is creative and thoughtful.

I hope that helps. Thank you for your input. I’ll scour the forums more to try and find similar issues/solutions.

Any categories you recommend in particular that I follow? Or even outside sources.

Simply the way AI models work: the more text they have as input, in this case a conversation, the less they follow the overall plan laid out in the instructive part of that.

Attention mechanisms have to see if more and more text is relevant to the user input as the conversation grows. Chat AI models powering ChatGPT also are overtrained on just being like ChatGPT, and the consumer version that treats GPT instructions as “only guidelines of what the user might like” does not offer as much reinforcement of behavior.

Start a new conversation and see if the GPT quality isn’t refreshed.