Mayan AI output behavior

Dear Mayan team:

I’m using the Comparative Comprehension plugin of Mayan Cognition to extract a summary of contextual matches between contract documents and using the output to run some workflows and handle the contracts. At first glance, it all works well, but when running the AI plugin over the same sets of documents, the results differ on each subsequent iteration. Is this expected, or could this be a bug?


Mayan Cognition is an experimental closed project and not part of the open source Mayan EDMS. Please use the appropriate channels when discussing these projects.

LLM are generative in nature and will produce different output on the same input for each invocation. This is not a problem with Mayan but a side effect of how these types of CNNs work. This disclaimer is presented when adding a cognition pipeline segment.