I’d like to share an experience I’ve recently been having with ChatGPT-4 that might resonate with some of you. I work extensively with Python, AWS, and other cloud-related technologies. In the past, I found GPT-4’s interface to be quite effective in handling code snippets. I could paste my code, and it would maintain context throughout our conversation, providing valuable insights and assistance.
However, lately, I’ve noticed a change that has affected my interaction with the model. After pasting the same code that used to work seamlessly, GPT-4 seems to lose context immediately. It often asks me to share information about the code I’ve just pasted, which can be frustrating, especially when working on complex projects.
[As a brief]
- I pasted the code with a question at the end (I’ve always worked like this with GPT)
- It ignores my question and proceeds to explain the code.
- Then i pasted the question again.
- Then it invents a piece of code that is not present in the code and claims it is a part found in my code.
- I repeat the question once again
- It apologizes and states he has no source code to look into
- Then I tell him the code was in the first message of the conversation, and he states that it is not true, as the first message is: “Proceed to copy the second message I posted, the one with the question only”.
- Then I pasted the source code again and began doing some tests on how much comprehension it has.
- At some point, it will start inventing code again.
Note: I know about tokens and all of that. My issue is: why did it work before but now it has become frustrating?