Why is the AI Meeting Summary So Inaccurate?

[Pamannmarketing]

I was so excited to try this, then so incredibly disappointed at the performance of it. It is WILDLY inaccurate. Like, not even close to the reality of what was said. Without any ability to train this AI, how will it ever get better? I always click on the thumbs down icon in the meeting summary emails to try to communicate to Zoom how bad they are, but it feels fruitless. What is Zoom doing to try to make this feature actually usable?

[Bort]

Hi @Pamannmarketing 

Sorry to hear that the Meeting Summary quality isn't up to a useable standard for you. We're working on various aspects to improve the language models, but it would be helpful if you could provide examples of meeting summaries with poor quality, so that we can look into these, see what happened, and determine what we can do to improve summary generation in such situations. 

Understanding your meetings may contain sensitive information for your organization, I'll reach out to you via  via direct message here within the Community for more information. 

[DG17]

Yes this is unusably bad at times. Please improve the accuracy quickly or this is not usable product. How can you trust these transcripts when the accuracy is not high. People will get incorrect summaries which is more harmful than no summary 

[VeronicaB2]

I'm also having the same issue. I've found the summaries and action steps by Fathom to be WAY more accurate, but I'm already paying for Zoom and don't want to have to pay for another product. I can't even share the summaries and action steps with participants, that will wind up causing more confusion than being helpful. As an example - in the summaries, Zoom is taking minor comments that were made and implying these were major points during a meeting. It's summarizing something that someone said in a way that actually changes the meaning of what they said. For example - someone said "I've found that when I involve x other people in this activity, I then get more done/done better" and it changed it to "this person said they involve other people to do the work" which is totally different, those people aren't doing the work - it's just that by integrating them in, it's causing others to do their own work differently/better. It's a subtle difference but means a lot in what someone might take away - and Fathom seems to get these details right, while every Zoom summary I've gotten so far has been so bad I can't use it. Also the next steps are WAY off - for example, in one the very first "next step" is that someone will open a Google doc so we can use it for brainstorming. That's not a next step, it's literally what we did in the meeting itself. It also somehow decided one participant is going to develop a model for a new service offering and offer it to clients - that in no way was what was said in the meeting, no one is developing a new model and to send something out suggesting that would be ridiculous and cause issues, not clarify anything. Completely unusable so far on my end. How quickly can Zoom get a product going that will actually be workable?