Hello Everyone,
I am trying to build an app that summarizes meetings and take action items. Which prompt do you guys think work the best.
E.g Meeting:
David: Hello Everyone, Lets discuss our weekly progress.
Jacob: Sure!
Peter: Sure!
Jacob: We had sold 10k of products
David: great! lets catch up at 5pm to discuss
E.g of possible output.
Weekly progress meeting
Jacob sold 10k
Catchup for david and jacob scheduled at 5pm.
possible prompt that i could use for getting this output.
response= openai.Completion.create(
model=“text-davinci-002”,
prompt=“Summarize the meeting and extract action items,
the meeting above /
Subject:/
summarize:/
Action items:”
temperature=0.7,
max_tokens=750,
top_p=1)
Do you guys have any other ideas that might work best for this use? or a better prompt?
Thanks!
I’m wondering how you will fit an entire meeting into a 4000 token (3000 word) prompt?
1 Like
Maybe use some open source hugging face sumarization or filter tool before handing the reduced prompt to GPT-3?
This might do the job for you:
Summarizes text of any length - 100 API calls / month free to test.
cody
November 2, 2022, 1:17pm
5
Take advantage of GPT-3’s “text competion” nature. Instead of asking it to do something (like summarize the above) act like you’re passing in a fragment of a document that includes a summery. This got me pretty good results with Curie:
1 Like
True I was originally thinking sending it by pieces but i don’t think that will work
So first i should summarize and then ask for action points? or maybe i missunderstood.
cody
November 4, 2022, 1:01pm
8
I just meant phrase/format things more like a fragment than a back and forth, so things like “Here’s the trasncript: … Here’s the Summary:” Instead of “Summarize the transcript above.”
oh ok thanks you very much!