90% of my process is training an AI model to be the world's best possible lawyer for one specific task

Then 5% is the AI executing and last 5% is me looking at it once, changing a few sentences, then validating the final output with a panel of other trained AIs to do final iteration.

Without doing the upfront coding the iterative stage used to take 40-60% of total time to finish. Now it's down to 5-10% with significant pre-builds

(http://www.autoadmit.com/thread.php?thread_id=5860436&forum_id=2)#49842162)