It’s just how machine learning has been since ever.
We only know the model’s behavior by testing, hence we only know more or less the behavior in relation to the amount of testing that was done. But the model internals has always been a black box of numbers that individually mean nothing and if tracked which neurons fire here and there it’ll appear just random, because it probably is.
Remember the machine learning models aren’t carefully designed, they’re just brute-force trained for a long time and have the numbers adjusted again and again whenever the results look closer or further away from the desired output.
I’ve been reading the book “A Small Matter of Programming” which discusses a bit end users relationships with computers.
I think people who are into computers get surprised to know most people just don’t care about how computers work and they shouldn’t have to. They want software that is easy to use and allows them to complete their task. Ex: a spreadsheet is an incredibly powerful software that hides anything about how computers work but still allow users to create multiple different “apps” by effectively programming.
This is one reason Apple is so successful and a lot of tech users don’t understand it. Apple creates “abstractions” so that end users don’t have to deal with low level details — something they don’t want to. They want to see the machine as a black box that just provides them some service easily and smoothly.
Most of the “decaying” tech skills people say are actually stuff people don’t need to know nowadays. Everything is an abstraction anyway, and most people tinkering with desktop computers aren’t aware of how the graphics software is rendering the screen, for example.
I think people don’t yet grasp that LLMs don’t produce any novel output. If that was the case, considering the amount of knowledge they have, they’d be making incredible new connections and insights that humanity never made before. Instead, they can only explain stuff that was already well documented before.
Yeah. I’m not sure that statement applies. It’s easier for humans to check something than to come up with something in the first place. But the thing is, the person doing the checking also needs to be proficient in the subject.
That’s true, but this happens because usually 95% of people are always on the latest version a few months after the new version was released. For developers, it’s really not worth supporting older versions when the overwhelming majority of users already upgraded.
Still, many large companies still support older versions when the user base is very huge. I work for a huge bank and we had to support all the way to iOS 10. Only this year it was recently upped to iOS 14, which now covers probably 99.99% of users.
It’s just how machine learning has been since ever.
We only know the model’s behavior by testing, hence we only know more or less the behavior in relation to the amount of testing that was done. But the model internals has always been a black box of numbers that individually mean nothing and if tracked which neurons fire here and there it’ll appear just random, because it probably is.
Remember the machine learning models aren’t carefully designed, they’re just brute-force trained for a long time and have the numbers adjusted again and again whenever the results look closer or further away from the desired output.