I have an all day meeting about to start - it’s been stated that Pocket doesn’t handle memories over 3 hours long well (or the AI models don’t)
I’m thinking I need to start stop memories to average around 2 hours..
Then - would it work to merge them? Not quite sure how to keep everything together, to keep it usuable… or just keep them separate and combine transcripts / memories after the fact?
Would love ideas - thoughts etc on how to approach a use case like this. Thanks!!
Hey @Sassa if it was me, I would do what you suggested, start/stop maybe every hour (if possible). Maybe even stretch to 2 hours as a test and see what the outcome(s) would be. Following!
I would like the Pocket team to weigh in - specifically around the feasibility of using the ‘merge feature’ after the fact. User shorter / longer recordings? etc..
hey guys! the best performance comes with recordings just under an hour from what we have noticed (in terms of speaker splitting etc as well).
we will also be introducing the custom dictionary but from benchmarking it performs best on meetings under 55minutes in terms of quality of output.
we have tags coming in as well so you can tag these recordings as well!
@mhirrrr - and thoughts about merging multiple one hour meetings to get a cohesive overview? how many merges before degradation of content? is there a limit or a threshold?
I would suggest using the global chat and as tags is rolled out today we will implement tag filtering in the global chat as well. we do our best to manage quality but llm’s do not do well with large context
@mhirrrr - yeah I was wondering about that - sounds good - will see how it turns out