...
All AI tools are imperfect and new. Consider the tools like you would an imprecise assistant; you are ultimately responsible for how the AI tools are used and for the content they produce.
Meeting analysis and summary is based primarily on the transcript Zoom creates. While word capture might be accurate, meaning analysis can be incorrect. We got better results when we treated the ZOOM AI companion as another participant in the meeting.
Next steps or action items are recorded accurately when declared. For example, “Next steps include Jane’s confirmation that the integration is finished; Emma will follow up with a status report”
For meeting video analysis, you must RECORD the meeting. Store the meetings in the cloud rather than on your computer (video takes up lots of space). (SU default? Zoom will delete recordings after 180 days.)
...
Meeting summary was not always accurate. Even if it captured the words correctly, the AI missed the meaning of the conversation creating a confusing summary that needed to be edited for clarity.
The tool has misgendered people even when pronouns have been set. It can misinterpret tone ie “Laurie expressed frustration.” We have reported these bugs to ZoomPronouns: During the pilot, ZOOM AI misgendered people. For accurate pronouns, set them in the Zoom Profile. If you do not set pronouns, the ZOOM AI language model assumes based on your first name.
We found ZOOM AI made other inaccurate assumptions. Always read and edit the meeting summary before sharing it publicly.
Speaker attribution only works when all users are signed in to Zoom accounts. Using an OWL or other conference software will eliminate accurate speaker attribution.
If you turn on the feature to “Automatically start Meeting Summary for all meetings I host”, the setting is NOT applied to meetings you scheduled before turning on that feature. This is a temporary problem that Zoom plans to fix. For meetings scheduled before the setting was activated, you must turn on Start Summary when the meeting starts.
...