Had another thought around the mediocrity thing … your poetry example is a good one. Someone who knows Yeats and good poetry would get that it is not as good as the original.

The point of ‘peak mediocrity’ is that current LLMs are good enough to be acceptable to more than half the population more than half the time (Not a high bar given our current state … but moving that aside ) and that ratio will only get better.


The old measure was that AI had to ‘beat the best!. Now you only have to ‘beat the average” … and the old meaning of ‘beat’ was ‘**win**’ … good, better, best in this new world is not absolute.

‘Mediocre people’ won’t be able to tell the difference .. case in point … your Yeats example … count me in as someone who would likely be ‘hoodwinked’.

Expand full comment

Lovely to see your words flowing across the screen John. Thankyou for taking the time to write - and I am totally with you. A big takeaway from Rao's points for me is that the AI challenges used to be 'be the best' - can you win Go, can you win Chess ... etc ... Now the challenge for the AI is to be 'good enough'.

Expand full comment

Mediocrity is not necessarily pejorative. For a longtime (since 1945) computing was based on Von Neumann architecture of function and logic. It was built by hand. A lot really sucked, but some was "insanely great."

Around 40 years later the notion of the software factory emerged because it took to damn long to make software. Lots of news techniques like OOP and new languages emerged. It was the dawn of mediocrity, or Von Neumann software that was just good enough, ordinary, suitable, and relatively easy to improve. However, it still was hand-made, expensive, and took too long.

In Europe for five centuries (11c-16c) associations of craftsmen called guilds made most of the stuff. In the 16c technology started to change rapidly, knowledge fluidity increased, and markets formed that allowed and encouraged trade outside the guilds. Over the next three centuries innovation diffusion increased, competition drove down prices, and products became interchangeable. Was it the start of mediocrity aka middling quality, not good, not bad, just right, good enough?

Meanwhile, at the Bugatti Molsheim Plant in France, 20 employees, taking six months, build the world's most powerful and fastest production car, with 1,800 individual parts, solely by hand. The Bugatti Chiron is not mediocre. While simultaneously, in Fremont, California, the Tesla plant produces 1640 new Teslas per day, every day, for an annual run rate of 600,000 new vehicles shipped per year. Is Tesla Model S mediocre?

I've used ChatGPT and Bard to write poetry. It ain't William Butler Yeats by any stretch, but its pretty good, quite close, usable and fun. Thing is the mediocrity claim often means playing the zero-sum-game. There's an old saying, ""Close don't count in baseball; close only counts in horseshoes and hand grenades." If LLM and AI computing is "close enough" today, if computing is mediocre today, then the chances that it will get better very fast, with little effort, are excellent! (That's another immutable characteristic of mediocrity.)

Expand full comment