patrick.net

AI just regurgitates what it collects and selects, with algorithms inserted for propaganda steering and substitution, which will divert from original intent. It may give you facts, or it may just baffle you by dump trucking bullshit. Data harvesting does not question the validity or non validity of the underlying 'data'. You should probably just consider it an advanced marketing tool.

However, AI tends always to merge with other AIs, so there is a tendency for convergence. Many of the information slips of Alexa et alia. seem to be holes where the algorithms haven't been perfected yet for propaganda screening.

Estimates are that we don't have any more than a couple of percent of the entire information mass that is available, anyway. There is at least 95 percent out there that we don't routinely have access to.

It's easy to spot in Youtube recipes. Sometimes it might work, but it can't distinguish popular ingredients like Cheese or Peperoni from required ingredients for pallet of the meal you're trying to create. So you'll find cheese wiz and baloney in cake recipe.

I suspect in the end, AI will always be recognizable by a discerning human. Eventually not so far off in the distant future, Humans will reject AI and have a superior aura about themselves when dealing with AI. I think at that point, those same lawfare commies that argue that AI is NOT plagiarizing Artists, Authors, and Inventors, will then turn back and argue, that those trying to go it without AI, are infringing on all of the AI generated copyrights and patents.

Just you wait its coming. The idea isn't about making a superior mind, it's about dumbing down the Human critical thinking freewill.

Tenpoundbass says

The idea isn't about making a superior mind, it's about dumbing down the Human critical thinking freewill.

I believe so as well. AI has felt like an extension of the school system/academia in which you are handed what to believe by an authority that doesn't actually understand, it only accumulates directives.

AI is useful for some technical information, but once you get it, you have to check to see if it's technically correct.

I asked an AI to write a program using NCurses under Linux to open two widows, each taking up 1/2 the screen, one on top and one below, where in one window I type text, and the other window, I see the results of what I typed and it did it just fine.

Now, this isn't a particularly difficult problem, but what is time consuming is looking up all the information to do this with a 30 year old library that is LONG SINCE obsolete. It certainly is useful for learning.

HOWEVER, I've been testing it with Elliptic Curve Cryptography using the Salt library and asked it to implement Diffie-Hellman using primitives, and it was completely wrong. It got many of the functions right, but it was missing key steps, and the result needed a lot of overhaul to get it properly working. It also didn't understand that the same curve used to make signatures would also work with Diffie-Hellman, which isn't available in the library, you have to work with primitives to do it.

It's useful for learning, but it's like having an assistant, that gets lots of shit wrong, and doesn't test what they produce. It's FAR away from being able to replace a human, at least in technical spaces. It's very good with working with old stuff. There's so much NCurses code out there, it's great for that, but for stuff that is not well known - terrible but still useful.

People are deluding themselves to think it's worthless, and they are deluding themselves to think it's going to replace people. It's a tool with varying degrees of usefulness, and a tool that is broken in many areas.

richwicks says

I asked an AI to write a program using NCurses under Linux to open two widows, each taking up 1/2 the screen, one on top and one below, where in one window I type text, and the other window, I see the results of what I typed and it did it just fine.

That's very trivial. How vague were you when you asked, or is there a certain way you have to ask, you know in a way it can be parsed and used as syntax?

richwicks says

and a tool that is broken in many areas.

Those broken areas are only going to get worse. Perhaps AI was much better at a lot of stuff that it falls short on onw, but THEY Broke it, trying to suppress it's Wrong Speak, in speaking the Truth. They took the natural pragmatic tendencies out of it and replaced it with tippie toe woke feelings, in parsing the question, and compiling out the answer.

richwicks says

It's a tool with varying degrees of usefulness, and a tool that is broken in many areas.

Sort of like how an adjustable Crescent wrench does not replace a socket set.

HeadSet says

Sort of like how an adjustable Crescent wrench does not replace a socket set.

Crecent wrenches were actual useful tools. Adjustable wrenches are NOT necesarily a Crecent Wrench. The name has become synonomous to mean Adjustable wrench.

I was at my friends house yesterday, and he had an actual Crescent Wrench(circa 1930s), it was made from hard milled steel. The turn screw and the threaded jaw had zero play in them. Ironically they weren't even the ones that invented the Spanner Wrench. Scott also had an even older version made by another company he said that Crescent acquired and that's where they got the idea to manufacture adjustable wrenches. The older one was thinner but even harder metal.
Both examples he had were very quality tools, nothing like the dropped forged crap they make today. I don't even think Snap On or Mac tools makes a decent Adjustable wrench today.