You can feed GPT3 a transcript and ask it "What happened in the meeting" but you can't ask "How was the meeting?"  It doesn't know how the interviewer felt, if the sales pitch went well, or if the meeting’s leader got what they needed.  It doesn't know the context outside of the meeting, it only knows the linear series of events.
‍
Of course, there are companies that are trying to achieve this magical ideal of automated meeting notes using large language models (LLMs).  We’ve tried these systems and they often fail.  They fail because A) many meetings aren’t recorded and/or B) they don’t have the right kind of information.
‍
I had a mentor teach me that meeting notes are meant to “synthesize the themes, bring in external pieces of information, build a narrative, and make it usable.”  That requires context and judgment.  Your automated meeting notes may tell you what was said, but it needs a human to create context for why it was said and what to do with it.  AI-generated notes need the right topics or questions, human input on those topics, and AI magic.

The right questions at the right time in the right place will help anyone create helpful context.  ‘What metrics were used’ might prompt a human to add links to those numbers.  ‘Why meet’ reminds readers of the purpose for the occasion.  ‘How might the session have been improved’ adds insight to the operational efficiency for the group.  Add context like this and you’ve instantly made the meeting notes worth reading!  Then, use LLMs to wrap it all into clear & concise writing.

What does all of this mean? It means that trying to align your team and organization is about to get 10x easier.