The guy that runs Nvidia was once a janitor. From a disadvantaged family, Jensen Huang attended private grade school in Kentucky from age 9 in the poorest county (then and now) in the country. All students had a work assignment. His was housekeeping. He told Joe Rogan he must have cleaned toilet thousands of times, adding that he had wished people were more careful. Though he didn’t say that the experience served him well, it plainly did. He is an unusually modest guy. He also confided that he was more driven by fear of failure than a need to excel. ‘I’m not an ambitious guy,’ he told Rogan. Nvidia is now the most valuable company in the world.
He also helped me understand AI. Rogan ran past him the recent case of the director who was going to disable AI. AI’s response when it found out? It threatened to go public about the affair he was having! (He wasn’t really. He’d just slipped it in as a test.) Rogan read ominous foreboding into this. Who wouldn’t? however, Jensen told him what had actually happened. In the course of its training, it had devoured narratives, perhaps novels, in which blackmailing schemes like this worked. So, he explained that the AI application has a string of algorithms regarding infidelity and a string of algorithms regarding blackmail and it just collated and compared, that’s all. Easy. He didn’t say ‘easy,’ I did, and it’s not easy from the standpoint of doing the math, but it is easy from the standpoint of knowing how the thing operates.
He reviewed the basic learning method of AI. The listener may ponder over whether it really is learning (as I did), but nobody will deny it gets results. What AI chips bring to the table is sheer brawn, sheer decision-making power. Break a task into the most minute steps imaginable, then break it down again into even more minute ones. Run the first of those tasks by AI, asking it to guess the answer. It will supply millions of answers, all but one of them wrong. Reinforce the correct one. Guided by this success, its next task will include not as many wrong answers. Reinforce the correct one. In this way, it gradually learns to “reason” correctly.
This is reassuring to someone who fears AI may usurp the spark of life that we thought can’t happen until God touches Adam’s finger, per the Michaelangelo painting. Not to worry. It’s just a huge numbercruncher. It’s “learning” won’t fool anybody, except for a few materialists who figure thats what life really is, a matter of numbercrunching to the nth degree. It’s like when Deep Blue beat Gary Kasparov, and some fretted that mankind’s goose was cooked, right then and there. Naw, someone else countered. Do you feel threatened at knowing a truck can outpull a man? That’s all it is, just transferred to the mental realm.
Don’t say this isn’t impressive. It clearly is. No way would I ever had foreseen it. To others who did, I was inclined to say, “What have you been smoking?” Make no mistake; it’s real impressive. But it still forever leaves that gap in being human. It’s like the limit concept in mathematics, AI comes closer and closer but never quite get there.
This explains why it will why it can consistently operate at genius level, then suddenly make a mistake that any two-year old would avoid. This is why, when I’m lazy, it can list eight 5-letter words that will fit what I’ve found so far in Wordle, only two of them are 4-letter words. ‘Oh, sorry,’ it says when called out. Then, upon being asked, it launches into a discussion of how LLMs learn differently from humans. ‘Anyway, here’s a revised list,’ it says, and provides another that also has two four-letter words.
It still has the remnants of being stuck by the question: “Are crocodiles good at basketball?” Although it could spit out any conceivable factoid regarding crocodiles and any conceivable factoid regarding basketball, that question would cause it to grind to a halt. Now it can handle the question with ease. Now it knows that crocodiles suck at basketball, but this only by running all the stats and finding that no team, from NBA to high school gym class, has ever drafted one. To reinforce this developing insight, it reviews data that good basketball players generally have long arms and compares that with data regarding crocodiles that generally have short ones. I mean, I’m oversimplifying here, as everywhere else, but hopefully you get the idea. It’s not really thinking. It still has no common sense.
Hmm, why does it not? muse the materialists, who will attempt to distill into algorithms what’s common about common sense. If only they could reach that point that they were enabling an Adam, they wish, and not God. Well, I don’t want to ever sell them short. But, in thinking they can digitalize the sacred through unlimited numbercrunching, somehow I’m reminded of that pop art “experiment” designed to test the scientific folk wisdom, “Supply an infinite number of monkeys with an infinite number of typewriters and one of them will write the complete works of Shakespeare. Infinite was not within the budget so they put one computer in an enclosure with six monkeys then awaited with bated breath to see what they would do. They didn’t write any Shakespeare at all; they shit all over the computer!
Uh oh. A new Nvidia chip is due, next-generational I am told. Jensen praises it up and down. Will this be the one that is like Dino, Fred Flintstone’s dog, that he puts out for the night, but then the dog sneaks in through the window to put him out?
****** The bookstore
Leave a Reply