I have a project with having GPT summarizing some conversation about activities, it does generally well but sometimes it seems to be having issue with dealing with tense: It tends to interprate future tense to past tense.
For example those terms mentioned in conversation：
I will go ahead and do …
If no objection, I will…
It’s ok to …
I will proceed to …
will be translated to I did …
Does anyone experienced the same, and is there any solution for this?
Your task is to summarize our company’s service calls by extracting the following information from the transcript in bullet point format
A) past activity
B) future action points
Give this a shot and if it doesn’t work better you should provide some re-written sample material and the prompt you are using.
temperature is perpetually stuck in “creative” or even “grammatically incorrect” mode, often producing unexpected word choices.
One could posit that a wide variety of outputs are useful to get thumbs up or down feedback for training.
For reliable output, giving you only the way the AI would most like to write, one can use the programming API, where you have control over these language parameters to reduce less-probable token choices.