GPT keeps referencing old files and ignores modifications to code during the past week

Understanding Persistent Reference Issues with GPT-Plus: A Recent Observation

Over recent days, many users have noted a recurring challenge when working with GPT-Plus: the model persistently references earlier uploaded files, neglecting recent modifications within the same chat session. This behavior seems consistent and affects the accuracy and efficiency of collaborative code development, especially in extended conversations.

Background and User Experience

Historically, GPT-Plus has demonstrated robust tracking of code changes during interactions, maintaining awareness of the “current state” even across lengthy discussions. However, recent reports indicate a shift in this functionality. Specifically, users have observed that:

• When uploading a file with instructions to modify the code (e.g., adding a new function), GPT-Plus responds appropriately by suggesting code snippets.
• Upon implementing suggested changes and informing GPT-Plus of the modifications, the model often fails to recognize them in subsequent responses.
• Instead, GPT-Plus references the original uploaded files, ignoring the latest adjustments, thereby leading to redundant or conflicting code suggestions.

Detailed Workflow Illustrating the Issue

Consider the typical workflow that now leads to communication breakdowns:

1. Upload a source file and instruct GPT-Plus to incorporate a specific change, such as adding a new function.
2. GPT-Plus responds with the code snippet, including the requested modifications.
3. The user updates the code accordingly and confirms the change.
4. Despite these steps, GPT-Plus insists the change hasn’t been applied, referencing the original file content or suggesting re-insertion of code.

This pattern suggests that GPT-Plus is not effectively maintaining the “current state” of the code base within the conversation, even when explicitly informed of updates.

Scope and Consistency

It appears that this issue is predominantly affecting GPT-Plus’s extended thinking capabilities, while the standard mode remains unaffected. Users have confirmed that the problem manifests reliably, making it a systemic concern.

Workarounds and Recommendations

Currently, the primary workaround involves re-uploading the modified files after each change to ensure GPT-Plus references the latest version. However, this approach is inefficient and can disrupt workflow continuity. Explicitly communicating that a change has been made often does not guarantee GPT-Plus will recognize it, emphasizing a need for improved session state management.

Implications and Future Outlook

This persistent referencing issue underscores potential limitations in GPT-Plus’s current approach to maintaining context during complex, multi-step development processes. For users relying heavily on ChatGPT for coding and collaborative development, understanding this behavior is crucial for