Hacker News Reader

Comment by mikewarot | Hacker News Reader

AI is only a dead end if you expect it to function deterministically. In the same way as people, it's not rational, and it can't be made rational.

For example, the only effective way to get an AI not to talk about Bryan Lunduke is to have an external layer that scans for his name in the output of an AI, if found, stops the session and prints an error message instead.

If you're willing to build systems around it (like we do with people) to limit it's side effects and provide sanity checks, and legality checks like those mentioned above, it can offer useful opinions about the world.

The main thing to remember is that AI is an alien intelligence. Each new model is effectively the product of millions of dollars worth of forced evolution. You're getting Stitch from "Lilo and Stitch", and you'll never be sure if it's having a bad day.

View on HN

Control freaks have a serious issue with the incompleteness of an LLM. Everyone else is just ecstatic that it gets you 70% of the way there often.

> Control freaks

People who like repeatable results in their work equipment are control freaks?

I know, right? Software Engineers with their zeal for determinism. How dare they.

Or modern mechanical engineers getting all pissy about "tolerances." Look, we shipped you a big box of those cheap screws, so just keep trying a different one until each motor sticks together.

Two different use-cases.

I'm promoting an AI directly and am interested in its output: 70% of the way there is a marvel

I'm building a tool that uses AI to do some piece of useful work: 70% of the way there is good enough for an alpha release.

Also, is there a known deterministic intelligence? Only very specific computer programs can be made deterministic, and even that has taken quite a while for us to nail down. A lot of code and systems of code produced by humans today is not deterministic and it takes a lot of effort to get it there. For most people and teams its not even on their radar or worth the effort to get it there.

> For example, the only effective way to get an AI not to talk about Bryan Lunduke is to have an external layer that scans for his name in the output of an AI, if found, stops the session and prints an error message instead.

> If you're willing to build systems around it (like we do with people) to limit it's side effects and provide sanity checks,

I don't think that comparison holds up. We do build systems around people, but people also have internal filters, and most are able to use them to avoid having to interact with the external ones. You seemed to state that AI's don't (can't?) have working internal filters and rely on external ones.

Imagine if everyone did whatever they wanted all the time and cops had to go around physically corralling literally everyone at all times to maintain something vaguely resembling "order." That would be more like a world filled with animals than people, and even animals have a bit more reasoning than that. That's where we are with AI, apparently.

> Imagine if everyone did whatever they wanted all the time and cops had to go around physically corralling literally everyone at all times to maintain something vaguely resembling "order."

I don't need to imagine anything. I live on Earth in America and to my mind you've very accurately described the current state of human society.

For the vast majority of humans this is how it works currently.

The amount of government, military, and police and the capital, energy, and time to support all of that in every single country on earth is pretty much the only thing holding up the facade of "order" that some people seem to take for granted.

> For the vast majority of humans this is how it works currently.

No it is not. Like I said, everyone knows everyone has an internal "filter" on what you say (and do). The threat of law enforcement may motivate everything (if you want to be edgy with how you look at it), but that is not the same thing as being actively, physically corrected at every turn, which is what the analogy in question lines up with.

Reply on news.ycombinator.com