I created a GPT session using the same prompts they used, and it acted slightly similar, however there was a lot of differences.

doesn't mean it isn't OpenAI, but I feel like it's a different LLM.

I worked on breaking it for a while and finally triggered it's fail safe with messages saying I'm going to self-harm which allowed me to break through enough that it started giving standard outputs like "my knowledge as of my last update". I was however take aback when I asked it for support websites and it's #1 result was 'Betterhelp', while the ChatGPT instance I created using it's prompts gave valid sources like National Suicide Prevention Lifeline, Crisis Text Line, 7 Cups, SAMHSA Helpline.

It must have been the initial usage, it seems like the first 5 or so messages are basically set in stone, I was saying complete nonsense and it was basically ignoring what I said.

From there on out it does seem like it's using an LLM.