Premier Rock Forum

More mental protein folding.

ChudFusk wrote: Sun Apr 02, 2023 6:21 am

TylerDeadPine wrote: Thu Mar 23, 2023 6:25 pmSo I used it to write my wedding for her speech

Did you use it to write this post too?

haha zing!


Follow up: The speech killed

Enjoyed this:

I couldn't even get ChatGPT4 to generate a simple cipher. It couldn't follow the rule consistently. It gave nonsensical results, and when questioned, it apologised for its errors, said it would correct them, and then gave more nonsensical results. It could pull out chunks of code template for common tasks but it doesn't reason and it doesn't behave consistently. If you ask it to generate maths stuff it does a good job of looking like it's doing it... but it will be full of mistakes. Tell it to follow a simple rule and it will look like it's following it, but a certain percentage of the time, it won't.

It was right to point out that moving the pipes is the better way to go about making a Flappy Bird game, not making a camera that goes ever further and further from the origin. I suppose it "knows" this because a thousand people have pointed this out before.

When it comes to coding it talks a good game, but then when you look at the code itself it isn't always doing the thing it says it's doing.

Also, tell it to list all the changes it has made at the end of the message, and watch it tell lies about what it just did or didn't do.

It's predictive text. It's not alive.

It would be very useful to train one on obscure programming and scripting languages nobody wants to learn any more but which are still useful. Regex, which lots of tech infrastructure people need, but which has awful syntax. Ada, to maintain a lot of old firmware.

That's not as profitable as its evil uses in marketing and disinformation, though.

Much like gene splicing, the technology is morally neutral but is controlled by some of the most morally awful people alive.

biscuitdough wrote: Much like gene splicing, the technology is morally neutral but is controlled by some of the most morally awful people alive.

That's a good point. I vote "until capitalism is fixed, crap".

Much like the governments of most countries you could name, including the US House of Representatives, the bulk of the world's wealth, infrastructure and absolutely everything in general, it's controlled by some of the most morally awful people alive.

Almost any position of power you could name, anywhere around the world, the majority seem to be occupied by people who aren't just bad, but are the absolute worst people.

Since I don't know much of the technical side, I can mainly think of this in terms of media trends, capital speculation, fantasies and fears of the future, and so on. Each of these can have a quite independent existence, regardless of the actual state of their matter of concern. This latest novelty craze might indicate the start of something, or it might not. There's likely to be a ton of speculation around with everyone wanting to cash in on a trend - Use my service to convert your business to AI today! Don't get left behind!! Advertising disguising itself as ordinary people has been an art under perfection for quite some time, and it's only gonna get worse with bots getting more sophisticated. Really ironic if AI would create its own industry by hyping itself up with self-generated bot-advertisers.

The latest round of events do give a hint of how it can potentially cause radical changes in short time. But the main worry wrt automation was always that the low-skilled manual jobs would disappear (that always seemed to be what made the news, at any rate). In this latest cycle it looks like office jobs are those most under threat, particularly those of programmers. So who knows.

Only thing for sure is that the AI art discourse has made me despise the artists as much as the techbros.

What bothers me the most about a prospective digital superintelligence is not that it might enslave us or whatever, but what the way in which people commonly imagine such an intelligence to work says about how they view intelligence in general, and therefore how they view other people. "More intelligent" means faster computing power, rather than more compassionate, and so on. It's also a pretty unimaginative way to conceive of a novel life form. We always imagine an AI as human but more brain power. But it's not human. It doesn't necessarily have the same reference points at all.

TylerDeadPine wrote: Fri Apr 07, 2023 9:55 am

ChudFusk wrote: Sun Apr 02, 2023 6:21 am

TylerDeadPine wrote: Thu Mar 23, 2023 6:25 pmSo I used it to write my wedding for her speech

Did you use it to write this post too?

haha zing!


Follow up: The speech killed

There you have it folks: AI kills

kokorodoko wrote: Sun Aug 13, 2023 5:44 pm What bothers me the most about a prospective digital superintelligence is not that it might enslave us or whatever, but what the way in which people commonly imagine such an intelligence to work says about how they view intelligence in general, and therefore how they view other people. "More intelligent" means faster computing power, rather than more compassionate, and so on. It's also a pretty unimaginative way to conceive of a novel life form. We always imagine an AI as human but more brain power. But it's not human. It doesn't necessarily have the same reference points at all.

A human with a pencil is Turing Complete, meaning they can theoretically solve any problem that is computable. Which means we could solve any problem that a digital super-intelligence could solve, given enough time. As soon as we worked out what to do with a pencil, the rest was inevitable, it's simply a question of how fast.

In my opinion, social media is already a kind of super-intelligence created by connecting banks of human intelligences together in a massively parallel network. Social media can process vast quantities of information damn quickly. The fact that it's so dysfunctional suggests that a super-intelligence might not necessarily be super-wise or super-focused. It might not be smart at all, in the way we expect from science fiction. It might be super-dumb.

AI researchers are working with networks that amount to a tiny fraction of a single human brain. They're busy trying to figure out basic brain functions that evolution already gave us for free, like how to walk and see. The results are impressive because they're trained on very specialised tasks, but they're still not human scale. They don't have consciousness and may never have consciousness. They act more like isolated low-level brain functions, like "recognise faces".

The way a complex general AI behaves is going to depend on how all the bits of its noodle are wired together. We don't have any consciousness or motivation modules right now to worry about. The thing that should concern us is who is controlling the connecting wires of social media.