We tried GPT-3 for our email startup (Mailscript), initially as a fancy way to detect and understand the content of an email. That didn't work out great, becaus

We tried GPT-3 for our email startup (Mailscript), initially as a fancy way to detect and understand the content of an email. That didn't work out great, because it's really prone to false positives and ultimately requires more work than doing fancy Regex. We're hopeful it will solve other problems we'll run into though, but we're not going to push for it before we find the problems. I'll share the lessons learned from the implementation if that's interesting to people here.