🔒 OpenAI/GPT: How does the web client refeed prompts, responses into its context?

[quartz tiger]

As per title. I notice that the chat often bugs out when I enter prompts with context >17,000-20,000 characters (clear that it has hit its context window limit).
However, if I divide said prompt into chunks and name them accordingly, GPT is able to reference them correctly.

Example:

[Long_Prompt]

User:
Answer my question on the essay below.
Lorem ipsum dolor sit amet, ... [20,000 characters long]

GPT:
// unable to receive the instruction to answer the question, instead takes cues from within the passage

[Dissected_Prompt]

User:
Label this as [Reference_A]
Lorem ipsum dolor sit amet, consectetur adipiscing elit... [10,000 characters long]

User:
Label this as [Reference_B]
Duis aute irure dolor in reprehenderit in voluptate velit esse cillum dolore eu fugiat nulla pariatur.  [10,000 characters long]

User:
Now, with reference to [Reference_A], [Reference_B], answer my question

GPT:
// answers successfully

This 'labeling effect' is true even for GPT's own responses. Eg. 'Your previous response suits the explanation, label it as [Reference_C].

I have done this several times but do not fully understand what OpenAI is doing on the backend that allows this. Help?

[compact trout]

@quartz tiger

Python help channel opened

Remember to:

• Ask your Python question, not if you can ask or if there's an expert who can help.
• Show a code sample as text (rather than a screenshot) and the error message, if you've got one.
• Explain what you expect to happen and what actually happens.

:warning: Do not pip install anything that isn't related to your question, especially if asked to over DMs.

Closes after a period of inactivity, or when you send !close.

[compact trout]

@quartz tiger

Python help channel closed for inactivity

This help channel has been closed. Feel free to create a new post in #1035199133436354600. To maximize your chances of getting a response, check out this guide on asking good questions.