Want improvement in chatgpt

i told a simple thing to chatgpt to not to reply when i type “suno” but its seems there is compulsive function of replying to every chat. it obvious it can clearly remember previous chat and has intelligence to remember and respond to them but would be nice if its not compulsive behavior’s of chatgpt to respond if command given to not respond it will make this machine more human like thank you team chatgpt

The entire point for an LLM (or any AI for that matter) is to produce an output.
You can learn to affect the output (the response), but you can’t prevent it from producing an output when you give it an input. That makes no functional sense.

2 Likes

well i gave it an input not to respond when typed “suno” and the function was not performed clearly that is where i want improvement. Thank you

This is an interesting idea and one I’ve thought about occasionally when I wanted to put an amount of text into context that was too large to be allowed in a single submission in ChatGPT.

What it sounds like you’d like to do is perhaps fine-tune a model to immediately produce the end of message token when some set of conditions is met, effectively producing a NULL response, which is interesting as a thought exercise.

I’ve gotten around this with instructions for it to only respond with “OK” until I send the message “I’m done” which was successful in getting it to not start producing an answer before I gave it everything it needed to know.

If you’re using the API though, I’m not sure how or why this is an issue as you can simply choose not to send anything at all to the model until you’re ready to have it respond.

Have you tried something along the lines of “if the user sends a message that consists of the word"Suno” then reply with " " only."

Since a blank space is still a token the model produces output but it should be a short and concise answer close to your requirements.

Otherwise the request to not answer amounts to the user not sending a message. So maybe it would make more sense to not send that message at all.

1 Like

Modern LLM’s are trained to answer any of the user’s messages, not performing their functions, so you can either use @vb’s advice or otherwise the functionality you are requesting is an app level, meaning that in ChatGPT you cannot do it, but you can do it if you are building an app using OpenAI API.

Don’t say cannot

https://chat.openai.com/share/9548bb09-9f11-4f0d-8879-858801070c55

(and in case you doubt the efficacy, inspect, copy, view in hex editor.

image

1 Like

This is a solution like @vb offered, not a solution to the @sejalplywood2802 's request.

When I post, I like to inspire the spark of an idea, but I guess if I have to do all the work…

https://chat.openai.com/share/4a656c9a-e83f-4561-a796-047a895dfa9c

1 Like

How is that solution the topic starter’s request?

i told a simple thing to chatgpt to not to reply when i type “suno” but its seems there is compulsive function of replying to every chat. it obvious it can clearly remember previous chat and has intelligence to remember and respond to them but would be nice if its not compulsive behavior’s of chatgpt to respond if command given to not respond it will make this machine more human like thank you team chatgpt

In your chat it is still responding (“” is not the lack of response).

I think we are running in circles now.
The main issue remains that if the user clicks “send” there will be a reply.

One either does not “send” in case it’s not wanted to get a reply, accepts a workaround as a reply or maybe manipulates the DOM to hide the reply if we are talking about ChatGPT.

1 Like

I would guess the ts is talking about as user facing GPT, so DOM unfortunately is not an option.

Guys the solution is pretty simple the programer who made chatgpt just has to include function of not responding when user gives it command. Currently it has a compulsive behavior of responding its so simple i don’t know why people are debating on it :joy: And making it difficult

1 Like

Well, then I have a follow up question:
Why are you requesting this feature?
While you stated what the goal is it is not clear what the actual purpose or value of such a feature might be.
I am sure it will be more likely to get such a new feature if the intended use cases are properly laid out in advance.

1 Like

Not logical. If you send it a prompt what do you suppose it should do with that prompt. It receives your prompt and must act on the prompt. Even if just to say ok. If it did not act on the prompt then what else could it do with the prompt? It doesn’t have memory it can store up prompts then act on them when you command it. It’s memory is the context window itself

1 Like

My friend when i command gpt to not reply on suno it clearly understand and in response its says it will not respond when the word suno is type but again it does respond to the word

Coming back to your follow up question it does serves a purpose if this function is included it will show that gpt is no more compulsive in its process the main point of making chatgpt was to make it more like human and we are humans only when we are not driven by compulsive behavior like other mammals

Still if you are not satisfied what can i do :joy: lets not drag this simple topic so much :slightly_smiling_face:

You fail to understand how gpt works. You need to break down the process into its basics. If you understood what was happening you’d understand why you’re request makes no sense.

1 Like

First, I think there’s a strong argument to be made that humans are pretty compulsive on the whole.

Beyond that, what you seem to be talking about appears to be more of a philosophical issue than a technical one.

You’re describing agency, the ability to make decisions and take actions for one’s self. I don’t see how bestowing the application with an instruction to not respond increases its agency—it would still be following your instructions, just to not respond in this instance.

From a user perspective, what would be the perception of an AI agent who could choose not to respond? It would appear broken.

Going further, once an AI agent does possess agency, we get into very tricky ethical territory that will be the domain of philosophers to navigate. Concerns about AI alignment become even important at that point as well.

Remember, at this point we are working with rather simple AI tools, not AGI or ASI. If your interest is in creating a simulation of AGI with the ability to simply ignore a user, that can certainly be built using the API if you wanted and it would certainly be an interesting novelty, but I don’t think it would have much utility beyond, perhaps, helping teach a certain subset of users appropriate social skills.

Something like the initial rollout of Bing Chat where the model would go off on users who insisted on calling it the wrong name and abruptly end the chat, refusing to engage further. That was something I genuinely enjoyed seeing and I wish the model still had the ability to directly tell users when they were being terrible.

3 Likes

I’m going to go in a whole different direction in answering this topic:

In ChatGPT, if you want to input multiple paragraphs without having the AI respond, you can hold down the SHIFT key on your keyboard, and then press the “ENTER” key, (or “return” or “suno” key if that’s what’s printed on your keyboard)

That key combination will allow you to insert new lines and paragraphs so you can keep typing in the same input box.

2 Likes

I think it would be a good improvement of chatgpt if it shows how many messages have you sent 'till now, I’m new using the Chat-GPT Plus, I’ve been trying a lot of things and I’m a little worried about message amount managing because you could easily reach the limit while working.