Fine tuning on when to create AI drafts
under review
Austin
Give rules and feedback on when to create AI draft vs. not create an AI draft. Right now it generates AI drafts in a lot of useless scenarios. It would be great to give the AI feedback and have it learn over time.
Examples:
* When I forward an email from my work email to my personal email, it's not helpful to have an AI draft written back to myself
* When I get a marketing email from a company, it's not helpful to have an AI draft telling the company's general/no-reply address "thanks, I'll take a look"
* When a web app notifies me of a new request or message on behalf of another user, it's not always helpful to draft an auto-reply to the app company (e.g. replying to notifications@linkedin.com saying "Okay, I'll add you to my network")
Duda Bardavid
marked this post as
under review
Austin
Here's another example of overzealous AI drafts. This kind of thing happens frequently.
Paraphrase of automated email:
We appreciate the opportunity to serve you. Please click the link to fill out a survey about your experience.
Useless AI draft:
Thank you for reaching out.
I appreciate [Company's] commitment to improving user experience. I'm happy to provide feedback.
Before I proceed, could you please clarify the following:
- What specific aspects of the project are you surveying?
- How long will the survey take to complete?
- Will the data be anonymized, and how will it be used?