My understanding was that the main issue is this would make spam too hard to detect. There is also the question of whether we are overanalyzing the output of t

My understanding was that the main issue is this would make spam too hard to detect. There is also the question of whether we are overanalyzing the output of the model. The model pretty much spits out garbage, but our brains are hellbent to find patterns in that stream, because thats what brains do. A simple analysis would probably show that it s no more meaningful than random words.