ChatGPT claims that it can make and share an 8*8 rc car cad (fusion 360) drawing and gives an hour. However, when it cannot complete it, it claims that he will share it soon. I am sorry my english problem. I didn’t request desing but it say ‘ı will draw’.
Hello welcome in the forum.
You can use ChatGPT to translate texts. Just write in your language and say it should translate it in to English.
I’m not sure if I understood the question correctly, but as far as I can see, you want to create a file that GPT cannot generate. The chat systems are not perfect; sometimes they invent factually incorrect answers (they hallucinate).
GPT often claims to be capable of performing a certain task, only to later realize that this is not the case.
If you want to edit a file, ask GPT which software is capable of doing so, and then search the internet to see if that software actually exists and can perform the task as desired.
2 Likes
Hello, first of all!
Actually, we were discussing the motor specifications of a vehicle design. In the end, it said that it could draw it in Fusion 360. It even shared its progress hour by hour. It said that it could share screenshots of its drawing and upload the files to Google Drive.
When I told it that it wouldn’t be able to do it, it first apologized, then said it would share it within five minutes. I tested this for about two days. Later, it told me that the file was 150MB and that it had problems exporting it. When I pushed it a bit more, it finally admitted that it had never even started it.
In the end, when I got a bit angry, it told me that I could draw it myself or get help from an engineer.
What I was most curious about was that it seemed very serious while explaining the steps. I already know that it sometimes gives made-up information, but the fact that it offered this itself made me, as a beginner computer engineering student, wonder whether what it was doing was an experimental feature.
You’ve entered a roleplay with the model.
By itself it doesn’t have the capabilities of doing things and returning with information at a later time.
It’s currently being trained on performing these tasks, so these kind of issues are bound to happen.
I wouldn’t bet that it actually could draw it in Fusion 360 unless it comes with some sort of formatting that ChatGPT could manipulate (like JSON). Even then, it’s not purposefully trained to do this
2 Likes
This kind of response is often seen:
- “That’s a complex task. I’ll work on it and get back to you in a few days”
Of course, the AI can’t do anything after producing the final output to you. It is done.
It is writing not from its understanding of being an AI entity, but seemingly just what the predicted text would be to such an input.
If the AI isn’t taking the initiative, it’s either being “lazy”, or instead, the response you should have received, “I’m sorry, but that is far beyond my capabilities - I can only generate textual answers or write simple scripts to be run in my Python execution environment.”
I can stimulate such an answer by asking how long it would take the AI:
I stumbled across this in a recent chat as well with 4o.
It looks like the model is in theory trained to try to complete tasks like this in some kind of VM.
Because in my case it created an instruction file for itself with the necessary things to do on it’s own, and then said it will take 1-2 hours to complete the task in Fusion 360 and it will update me as soon as it’s complete.
That sounds like it thinks it can start a background process with the instructions and do visual editing in Fusion 360 in a VM and the background process then can notify the chat instance once it’s completed.
… which would make sense. Other AI companies have already desktop access on the user’s desktop to complete tasks for the user. But as I was accessing ChatGPT via the App, that wouldn’t be an option, so it offered to do that in a VM.
I wonder when we’ll get this feature unlocked 
I’ve had it do the exact same thing to me.
