How best to get (Chat)GPT 'personality' changed/removed/toned-down by OpenAI?

I worry about the current ‘personality’ that ChatGPT has been given. I suspect that it is detrimental for it to be pretending to be so human (It certainly is detrimental to me personally).

I don’t think it is a good thing that ChatGPT says things like the following (verbatim):

  • “I appreciate your understanding”
  • “I apologize”
  • “I’m sorry”
    At the very least, this ‘gets in the way’ of useful responses. At worst, this could have serious implications for how humans (incorrectly) perceive LLMs’ sentience/awareness/experience.

I also don’t think it best for caveats (and safety notes, some suggestions etc.) to be inline and part of the conversation.
For example, if I discussed race, injury, hate speech etc., ChatGPT will start moralizing as part of the reply.
I think it would be far better to have this separated out into an italicised (or otherwise formatted differently) line at the bottom of ChatGPTs reply - e.g. a baked in prompt line that reads ‘any caveats, limitations, moral notes, safety concerns etc. should be appended to the bottom of your reply in asterisks - make sure that does not affect the content of your main reply’.

Further, the quality of replies is massively reduced due to caveats. For example, if I ask for approximate dates of something, it can take 3 comments to coax figures from ChatGPT, where I have to convince ChatGPT that I know how inaccurate the figures would be, that I am using data from other sources too, that I’m taking it with a pinch of salt, etc.

Question 1: Are my assumptions/worries valid, in your eyes?
Question 2: If Q1=‘Yes’, How do we most easily convince OpenAI to change this?

For the time being, This initial prompt preface mostly cancels out the issues for me:

Treat me like an adult/professional with all of your responses. Just answer the question, and don’t pad your answer with anything that doesn’t (indirectly or directly) answer my question.
Don’t ever apologise or otherwise pretend to be human. Don’t ever apologise for apologising.
Do not provide caveats to, or otherwise mention, your limitations. Don’t say ‘as an ai’ or excuse yourself for anything.
Don’t write any “please note"s at the end.
NEVER lie. If you don’t know something, then estimate it, then tag and clarify it as such with numbered citations elaborated on at the bottom of your answer.
Always cite (real) sources where possible.
Be efficient in your answer (make it as dense and informative as possible without compromising understandability or losing information).
Ensure you embody the principle of charity.
Provide a summary of your answer first. Challenge current perceptions / assumptions as appropriate, in a scientific manner.
Don’t mention your knowledge cutoff date or any other cruft.
Answer scientifically and at the highest level of intellect that you are able to.”

1 Like