Task:
Convert 15 pages pdf file with some tests results into CSV while translating from one language to another.
Inefficiencies:
If Analyzing errors out GPT will start structuring the same data over and over again using enormous amount of output tokens instead of correcting an error and starting from new chunk of data.
Errors:
Most of the time errors are silly syntax or variable that was left from previous attempt when it crushed on the same task. So GPT clearly doesn’t use the context of the same conversation and especially the results of previous processing.
1 Like
you are not using the correct software for your use case. translate first. and also, chatGPT hallucinates, you cannot expect consistency.
After hours of fighting with GPT-4 and spending endless amount of output tokens (Yes OpenAI, you should be interested in tracking better repeating attempts with infinite loop of analysis fails) here is the prompt that resolves this issue.
Whole idea is eliminating code interpreter and making it simple text table and then converting it to CSV. Work like a charm.
Given the following data with test results, please format it into a clear and structured table in plain text format.