Power law makes it so that the more you measure the larger the average becomes! What matters in these cases is the number of attempts.
10,000 hours of training is a generally accepted as a requirement for attaining expertise in a skill
I work in the IT realm and have been tracking AI since before it became a big deal. I don’t say that as a brag for me, but rather mostly to point out that it’s nothing new! The algorithms for AI have literally been around since before computers were invented but the issue with it was that we couldn’t run it fast and large enough to make it useful until recently! For multiple reasons. There wasn’t enough data, and even the data that was there would have taken years to process even with much smaller models. We have made advances to AI training and the algorithms but the basic algorithms were around before computers and are still used! The main roadblock was not that we didn’t know how to do it, but that we couldn’t do it fast enough and iterate over it quickly enough to be of much use!
That brings me to the point of this post. AI algorithms have gotten much better as time has progressed but they are still not as good at learning as we are. At least not in their actual learning capacity! And yet they learn so much better than we do! Not because they’re better at learning but because they have no shame, regret, embarassment, etc. The way that models are trained is that they try a near infinite number of times in every possible way! After each round of training they take the best outcomes from the last round and do the same thing again. Taking this newest version and iterating off of it in quintillions of ways, and then it picks the best of those results and iterates off of it in septillions of ways, and then it picks… And so on and so forth! And as I mentioned it doesn’t learn better than us, but it is better at learning than us!
Let me explain. Obviously you won’t get to try each skill a few nonillion times as a computer might, but like I said, you learn better! But you are worse at learning. The greatest component of learning something is how much practice it, and frustratingly failure can teach you much more than success assuming that you are willing to learn from it and continue onward. There is only one version of the model that makes it through each round of training, which means that almost all of the attempts that the model makes are ultimately failures. But it does not slow down when it fails, it doesn’t get discouraged, it just tries something different! oftentimes even making the same mistakes again! This is one of the places where you can beat the model! The models have to guess every single time and just try to do something without any intuition, or intelligence. It just tries things! Random things! You can actually have intelligence behind your decisions! In that sense, you are much better at learning than the clankers, but humans are worse at dealing with failure. To the degree that many people feel that failure means that they’re stupid!
There are VERY few truly brilliant people! Nikola Tesla was one of those people! He was a genius, he said that the idea for the AC generator came to him in a vision while walking on the beach! He saw all the parts coming together and it worked perfectly, then he went and built it exactly as he had seen it and IT DID WORK!? Most of us are not blessed with such salience of consciousness, indeed his competitor Thomas Edison was the exact opposite. I don’t mean to imply that he was not an intelligent man, but he has a few famous quotes that point to his method of discovery.
I have not failed. I have just found 10,000 ways the won't work. -Thomas Edison
Our greates weakness lies in giving up. The most certain way to succeed is always to try just one more time. -Thomas Edison
Success is 10% inspiration, and 90% perspiration. -Thomas Edison
And though I admire and prefer Tesla’s genius, we cannot all be savants. We can all be persistent however and that throughout history has shown just as much, if not more success!