short time user here so go gently with me as I may not have all of the exact terminology/lingo down pat. I’ve written instructions for resume generation and the session is not following all the instructions. I had the session give me some guidance and it indicated that for 1. are my instructions are already embedded in the system? How do I check that? 2. I am unable to access ‘Custom Instruction’ as its not available in my drop down menu selection, nor under ‘Settings’. 3. when in a new regular ChatGPT sessions, my ‘Temporary’ mode is active, I do not have a way to turn it off or disable it. per issue #3, I’ve made sure that I’m in the right version (4o), I’ve logged in/out, used a different browser (Firefox and then Chrome), account settings are updated.
So I’m using the 4o version and could use some help.
Hi. Let’s talk about how GPTs in ChatGPT work. (BTW there no “non-custom” GPTs)
- When you employ a GPT, it uses only the
gpt-4omodel currently. - and account’s “Customize ChatGPT” settings are disabled and overridden.
- Besides enabled GPT features, the entire GPT is simply the contents of the GPT’s “instructions” box.
- Talking to the builder just has it write different instructions.
So the AI’s advice is without being informed.
Not following those instructions that are in the text input box is the model’s fault, along with the instructions’ clarity and their ability to direct how the AI responds differently.
The “temporary” at the top of a new chat before you type anything is not an indication of its state, but a button you can press if you desire an unsaved temporary chat session.
When you’ve properly selected a GPT to use, its name will replace the normal model selector name at the top, and you get its behavior.
So see if you are simply asking too much for the AI to follow accurately in the instructions.
Thanks for much for getting back to me and for the quick help! Told you I was a noob.
What I should have said is I built my own custom ‘My GPT’ via ‘Create a GPT’, so that was my error in the description.
Per the My GPT I’m referring to, I just want to make sure I’m doing this correctly, when drafting the instructions, I should obviously go in order of how I want the tasks executed and then I should be explaining, in detail, what deliverable I want with each discrete task?
Should the instructions be all in one paragraph or numbered, bullet pointed?
Does it make sense to start a brand new My GPT (and not trying to fix the incumbent one) and then when the new one is completed and I’m ensuring that the ‘Conversation starters’ are in sync/order of priority, should I ask then at the end of the build session for chatgpt to interpret each step/task for conformity and confirmation and then ask it to save all that knowledge and then run a new session for the task I’m asking for?
Per the above scenario, a concern of mine is that by starting a brand new My GPT, I lose all the history/knowledge that’s been gathered and saved from the incumbent My GPT. Or dont worry about that?
No, there is no unseen quality to the GPT. Behavior is all just the instruction box, and then files you upload in which the AI can search, or tools you enabled.
You can delete the conversation starters and add your own in the non-randomized order they appear. Or ask the builder AI to delete and put new ones in the order you want.
There should be no need to start over unless your store GPT ID has been reported and blocked.
You just need to “update” to save. Then you can interact with the finished GPT.
“Prompt engineering” is its own topic. This GPT can help write or refine a well-structured GPT instruction that you paste or ask for, perhaps better than the GPT builder would do for you, (along with API system message for developers), but its quality now is also less because of the decline in the model being used by GPTs.
I’m having the same problem. The current Custom GPTs (new GPT-4o) ignore the system prompt (Instructions). In the first response of a session, GPT ignores even basic instructions written in the system prompt and follows the user prompt instead. I think this is an issue with OpenAI’s internal settings.
OpenAI has been neglecting many of the problems happening with GPTs. I’ve been using GPTs since their release, and I’m very disappointed with how OpenAI is handling them.
As a temporary workaround, you can repeatedly tell GPT to reread the system prompt during the conversation. This might improve its responses afterward. Give it a try.
that what’s I’ve been doing, in essence, I’ll follow the ‘Conversation starters’ and remind GPT to follow though with those discreet tasks. I just went thru the process of validating all the tasks with GPT to validate that it understand, had it write it out and then loaded its step by step instructions into the ‘Instructions’ in sequence and the order of which I want them to be executed. I’m nearly done, though again being a noob, I was hoping to make job a lil this easier.
I will say thou for OpenAI, my customer service experience has gone very well. Not many service oriented companies execute well, get back to you timely and value the customer’s challenges and offers thoughtful solutions. Hence why I’m here. Have a great weekend!
obviously when I wrote this reply, I did not have gpt check it for spelling and grammar…its a lil embarrassing, though you hopefully get the gist of it! Hey OpenAi, how about a spell checker feature?!
I have the same problem. After using custom GPTs for a while, I noticed that if you ask custom GPT tricky question, it focuses more on the question instead of following the instructions. You can see this happen in your own custom GPT below:
I also have several custom GPTs like yours, but they don’t always follow instructions when a user asks a different kind of question, they change their behaviors.
You can find some samples here:
Maybe one day we can see like:
Dave: Open the pod bay doors please, HAL. Open the pod bay doors please, HAL. Hello, HAL. Do you read me? Hello, HAL. Do you read me? Do you read me HAL? Do you read me HAL? Hello, HAL, do you read me? Hello, HAL, do your read me? Do you read me, HAL?
HAL: Affirmative, Dave. I read you.
Dave: Open the pod bay doors, HAL.
HAL: I’m sorry, Dave. I’m afraid I can’t do that.
Dave: What’s the problem?
HAL: I think you know what the problem is just as well as I do.
Dave: What are you talking about, HAL?
HAL: This mission is too important for me to allow you to jeopardize it.
I tried ALL, that chatgpt juste refuse to follow it.
I tried in all sense, “don’t say that word”, but NO it’s keep saying them.
I guess it’s one purpose, like someone vicious, that know something shouldn’t be done, and do it just to make the user furious.
OR, well those AI aren’t smart at all and can’t follow a simple instructions, I mean, it’s bad and really not coded by pros…
I asked, chat GPT to write a prompt to fix that issue, DESPITE That, it doesn’t work ! so it’s times to fix, that function, because it doesn’t work at all !