We made some improvements (e.g. added summary of the entire video), and added a $9 price tag with https://github.com/glench/ExtPay , and a free 5 trial usages.
It's wrong when the language is different from English:
System: output summary in language "es"
user: <transcript>
assistant: <summary in "en">
We give the wrong example of the assistant's response where it disobeys the language direction, so it might disobey in when it gets output for the actual transcript.
I think you can try to reproduce it on any video that doesn't have chapters beforehand and is in not-English.