Reuses transcriptions, new default model

Due to the release of the very capable LLaMa-3 model that's the new default. However, if you use this script for very long recordings the larger context window of Mixtral might mean you should still use that model.

The script also saves transcriptions and remembers which URL they're from, which means you can rerun the script in conversational mode on existing transcriptions which saves a lot of time.

Conversation mode added

This release contains all the originally planned functionality. With the conversation mode (-c) added to the commandline, the script will now wait for further questions after having printed the summary. This has turned out to work extremely well even on very long videos/podcast at an hour's length.

First public version

After having blogged about this little script a lot of people asked me for it. I've thus cleaned it up as is and here it is for your enjoyment.