Hey I’m trying to optimize a prompt where the user can put a few bullet points and topics for an email and then select a tone/mood to generate an email from
Create an email to a customer from the following bullet points and make it sound excited:
- You expressed interest in our product
- Want to be beta tested?
- Skip waitlist and get access
- Looking for feedback on product
- Happy to send over instructions!
the output I’m getting is
We’re so excited to hear that you’re interested in our product! We’re looking for beta testers to help us test the product and provide feedback.
If you’re interested, please fill out this form:
The outputs have been pretty good so far across a small sample size on the Davinci Instruct series with the following settings:
temp = 0.5
top p = 0.5
frequency penalty = 0
presence penalty = 0.5
Best of 1
Can someone provide some feedback on this prompt or how I could make it better? Getting the mood/tone correct is very crucial. I will test out with examples as well but another perspective would be awesome
Thank you for your concern but it’s not something I’m worried about. I have enough experience in this space and my focus is on optimizing this feature. Any help/feedback on the prompt is appreciated
Even as a human I can’t quite interpret what the bullet points are supposed to convey. Try using only qualitative natural language instead.
The use case is to generate emails in a specific tonality from bullet points. I could of course create a prompt in natural language and define the email output I want but that defeats the purpose.
As a commercial app I can’t have so many prompts for an infinite number of emails that would be different.
Check out otherside.ai for reference as it’s the same feature without a mood/tone preset on the output
As it may be, NLP features have always been used to improve email productivity (smart compose, auto reply etc) and there is a growing market need for even more email writing experiences in a professional setting.
How many people really enjoy sitting and typing out a meeting report summary email, or sending out investor emails/updates or replying/creating prospecting emails? This is not for grandparents communication with their grand children obviously.
Otherside and many other email based productivity apps have already raised institutional capital proving there is an inherent desire and need for this technology being applied to email.
I understand your position about it but I feel its slightly misplaced and not conducive to the idea - make cool features with GPT3. Email productivity and experiences are one of the best ways to showcase the text generation capabilities of this technology.
Gatekeeping technological development seems like an inherently bad idea.
You can dynamically compose a natural language prompt. But if you need to use lists the best way is to label them.
Hey, I like this idea…how would I adapt this prompt to include multiple topics/subtopics? Also, can you elaborate on how I’d dynamically compose natural language in this use case?
The basic benefit is the user can clear their mental bandwidth about writing the email by just noting the key points and intent for the email and then generate it.
GPT-3 is really good at generating ideas and lists. I use one prompt to generate ideas or questions and then subsequent prompts to fill in the blanks.