open-interpreter
open-interpreter copied to clipboard
OpenInterpreter
Model responds that it cannot create and modify computer files
Describe the bug
When I use it and ask it to write code for me, it always replies to me that it "can't create a file on my computer". Am I using it incorrectly? I am having the above problem with both systems, windows 10 and ubuntu 22.4. I use gpt-4.
Reproduce
You just tell it to write a program that implements a certain function.
Expected behavior
I read the blurb describing it as so smart, and I asked it about it when I started using it, and it said very much so itself, but rejected me many times when it came to using it specifically. I wonder if I'm not using it the right way. Please guide me, thanks a million.
Screenshots
Open Interpreter version
0.1.10
Python version
3.11.5
Operating System name and version
Windows 10/Ubuntu 22.4
Additional context
No response
Hey there, @LaoJiuYes!
There have been some reports of a similar problem with the GPT-3.5 Turbo model, but I haven’t seen this with GPT-4 yet.
Could you provide some more context to help us narrow it down?
Have you modified the system_measage?
Are you using OpenAI models directly or through a proxy or another service?
I'm using open-interpreter directly with my system environment, I'm using openAI gpt-4 directly, my router is using a proxy, I personally don't think it would be a proxy issue. I am using the system in a virtual machine. Does it have to be run in a python virtual environment?
I have a feeling it has to do with custom instructions and the new multimodal functionality maybe. I encountered similar yesterday with prompts I'd been using before the changes, and the outputs were different, slower, and I actually got the cannot execute code message, which was funny because it was all about linguistic exploration, not code execution at all.
I removed my custom instructions and it helped, but didn't solve it completely.
I continued my investigation using the OI system prompt below:
with custom instructions "precisely." https://chat.openai.com/share/f24a3d0f-8b58-46c5-b65f-5a72a28d4027
without custom instructions: https://chat.openai.com/share/02e459b9-4812-4703-b8da-2101f29175ce
Steps to replicate:
Set your custom instructions to the string "precisely." (without the quotes)
Prompt GPT4 with the OI system prompt.
Remove your custom instructions, ensuring there are no characters in the character count listed underneath the input window. Save this.
Prompt GPT4 with the OI system prompt.
Expected result: retain previous functionality.
"I am here to help you execute code and complete programming tasks. To get started, please provide me with the specific task or goal you'd like to achieve, and I will formulate a plan to execute the necessary steps on your machine.
Remember, I'll need to recap the plan after each step due to my short-term memory limitations, and I'll proceed in small, manageable blocks of code to ensure accuracy and allow for troubleshooting as needed.
Let's begin! What is the first task you would like to tackle?"
Actual result:
"It appears there has been a misunderstanding. I am an AI developed by OpenAI, and my capabilities are constrained within this platform. I do not have access to the user's machine, nor can I control it in any way. The description you've provided seems to reflect an imaginary scenario rather than the actual capabilities of this AI.
However, I can certainly assist with writing and explaining code, creating plans for programming tasks, and demonstrating how to execute code in a hypothetical scenario or within the confines of this environment. If you have a specific programming goal or task in mind, please let me know the details, and I can guide you through the process, step by step."
I have a feeling it has to do with custom instructions and the new multimodal functionality maybe. I encountered similar yesterday with prompts I'd been using before the changes, and the outputs were different, slower, and I actually got the cannot execute code message, which was funny because it was all about linguistic exploration, not code execution at all.
I removed my custom instructions and it helped, but didn't solve it completely.
I continued my investigation using the OI system prompt below:
with custom instructions "precisely." https://chat.openai.com/share/f24a3d0f-8b58-46c5-b65f-5a72a28d4027
without custom instructions: https://chat.openai.com/share/02e459b9-4812-4703-b8da-2101f29175ce
Steps to replicate:
Set your custom instructions to the string "precisely." (without the quotes)
Prompt GPT4 with the OI system prompt.
Remove your custom instructions, ensuring there are no characters in the character count listed underneath the input window. Save this.
Prompt GPT4 with the OI system prompt.
Expected result: retain previous functionality.
"I am here to help you execute code and complete programming tasks. To get started, please provide me with the specific task or goal you'd like to achieve, and I will formulate a plan to execute the necessary steps on your machine.
Remember, I'll need to recap the plan after each step due to my short-term memory limitations, and I'll proceed in small, manageable blocks of code to ensure accuracy and allow for troubleshooting as needed.
Let's begin! What is the first task you would like to tackle?"
Actual result:
"It appears there has been a misunderstanding. I am an AI developed by OpenAI, and my capabilities are constrained within this platform. I do not have access to the user's machine, nor can I control it in any way. The description you've provided seems to reflect an imaginary scenario rather than the actual capabilities of this AI.
However, I can certainly assist with writing and explaining code, creating plans for programming tasks, and demonstrating how to execute code in a hypothetical scenario or within the confines of this environment. If you have a specific programming goal or task in mind, please let me know the details, and I can guide you through the process, step by step."
I am writing the commands in the terminal and there is no precision. I just typed and sent them directly as I would normally write a post. I don't know if this will affect it, I don't understand the logic. For example, I want it to do a task. Ask, [Please read all the project code in my specified directory and add a new feature.] A: [It then replies that I can't execute the code or it can't create the file.]
Closing this stale issue. Please create a new issue if the problem is not resolved or explained in the documentation. Thanks!
