ML Researcher. I’m strongly interested in rationality, epistemology, as well as generative models, their shortcomings and societal implications. I also do music theory research on the side.
Currently at Meta.
ML Researcher. I’m strongly interested in rationality, epistemology, as well as generative models, their shortcomings and societal implications. I also do music theory research on the side.
Currently at Meta.
All of my school years I’ve been dreading assignments with minimum length. I ultimately always had to bloat my essays to meet the requirements. It’s sad to see that the same limits still exist today, like in some journals.
It’s hard to generalize, as some people may struggle to express all of their ideas when they don’t take the space to develop them properly. But setting a minimum will inevitably lead some arguments to be diluted in unnecessary prose for no good reason. Sometimes, one paragraph is all it takes to make a compelling argument.
I believe we have a natural tendency to conflate volume with quality: if you have a lot to say, surely you gave it a lot of thought. Obviously, this isn’t generally true. We find this bias everywhere, in long empty company meetings, in AI slop websites that could have been a short static page, etc.
I agree that the proposed ‘fleshman’ view point is quite abstract and very different in its use from the other two. It seems to me that the main way to make it not as abstract is given in the conclusion: as a means to predict what a group might do or how the group might evolve.
People can believe and follow good causes (steelmanned version of an idea) for partially wrong reasons (fleashmanned version). Debating whether the actual idea is good or bad is one thing, looking at what actions the followers of that idea will take is another.
In a way, given a distribution of beliefs in a movement, the strawman is the lower bound (easy to attack), the steelman is the upper bound (great to use to prove wrong), but the fleshman reflects the actual distribution of opinions between those two, which helps not debate the idea, but understand real world cases.
It’s easy to take a belief held by real people and straw/steelman it into oblivion to an extent that no-one in the group actually recognises the idea anymore.