How to stop numbered lists in output

Has anyone worked out how to always stop the output including a numbered list, I’ve tried
“Never create a numbered list” and similar, which works most of the time but occasionally out will come a numbered list

Hi david,

You can ask the model for the results in JSON format, or CSV, some known format like that, or you can use the functions system and define a function that returns an array of strings… should get you 99.9% of the way… never can say 100% with LLM’s or people!


Try in your prompt saying it for example 3 times just in different ways. I have used this in cases where he wasn’t doing what I asked it for and he started doing as I said. Sometimes I treat it like a child who needs to be told something couple of time and in different ways to finally understand what I’m saying. This should help as it worked for me.


Hi David are you could try specifying only output as a vertical list

1 Like

If you are talking about completions, its not very hard, ive never ran into a problem where i couldnt change the formatting. However, if you are talking about chatgpt (maybe you are working on a plugin) just tell it not to use markdown. The interface has a markdown renderer and when gpt starts using ticks it automatically renders them.

1 Like

I should clarify though, for completions just give a couple of short examples of allowed formats/outputs. It shouldnt give you problems. In fact, sometimes its uses the examples too strictly.

1 Like

Like others have mentioned, it would probably be best to specify the type of list you do want.

I know that at least for image generators, they can often have trouble when specifying the absence of something, as illustrated here, where asking it to create a plate with ‘no bananas on it’, it generates a plate with bananas.

So I would experiment with not even mentioning the numbered lists, because I’d wonder how much of the training data would be something like “here is a list that is not a numbered list” lol.

Very good point. Worth trying an example fewshot in your prompt, if you do not give an example of a list with no numbers that the model will use whatever sequence it happens to match on for the first entry in the list and then repeat the formatting as it’s been aligned that constancy of presentation is important.

presumably you could get away with something like

Do not do this 
1) this is a line
2) this is another line

this is a line
this is another line

LLMs are pretty poor at following negative instructions. I think it’s safe to skip the,

“Do not do this” portion and just give it two examples of lists it should do.