How to stop getting replies with Ai sounded bording words?

ChatGPT helps me in a lot of issues on daily basis, from formating issues, correction grammar, spellings. The most I use it for, is to Summarize a lot stuff, specially non-fiction articles online.
But I am tired of getting ai like words e.g. delve, dive, deep, (most common) tapestry, intriguing, holistic, intersection, not only … but … (standard inversions), dancing metaphors, Site-specific: (shipping) - navigate into, sail into the future(research) - ethical considerations, area, realm, in the field of (even if it already knows the field), in the age of, in the search of.

I don’t want to use API, to create custome GPT.

1 Like

Hi @HMAQBL !

That’s super interesting - there is a nice article in The Economist (paywalled) about some of these words that have a statistically significant occurrence frequency across LLMs.

One way to try and contain this is to personalize your ChatGPT (disclaimer: I have an Enterprise subscription, but this is available in free and other versions as well). So you go to SettingsPersonalizationCustomize ChatGPT and then in the How would you like ChatGPT to respond, you explicitly list the words you would like ChatGPT to avoid using.

Then when you start new chat sessions, it should behave differently. There are no guarantees however - you are essentially just providing an additional fixed-context that is prepended to all your messages, but in my case I have seen a difference in behaviour.

The advantages of the above solution:

  • You don’t have to keep repeating to avoid specific words in your chats - the behaviour is persistent
  • I noticed that just creating such rules direct in the chat doesn’t necessarily work; it seems that the “customization” part is a little bit like system prompt - it bootstraps the behaviour of ChatGPT.
2 Likes

Hi,

I know how you feel! I’m not into the word “elevate” at all!

Since you’re using the ChatGPT UI you have some different options: If you’re using the free version, just be very specific in your custom instructions. If you have a Plus or Teams subscription you can create your own CustomGPT (which is different from an API Assistant) to very specifically follow your wishes.

In all of these cases include a list of words you would prefer it not use, and examples of what you would prefer instead.

2 Likes

do any official from OpenAi could share insights about this?