Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[lemon10]

Fair enough, I left out a ton of qualifiers that I was too lazy to add (and because it would have been annoying to format ect).
As you say there are a ton of caveats.

The AI search engine are different on a few levels though.
1/2) They have access to more data then regular chatGPT due to being hooked up to bing/google, but fair enough, they only have access to public stuff and obviously don't recognize videos/games and probably not pictures either.
3a) They update much more frequently since their entire point is to know whats on the web, not quite sure how often, bing search either has current data and/or the ability to look information up in real time in its database. This problem at least is already solved.
3b) I mean, it could totally give you both sides, or (as already exists) give you one side that you ask for. Of course it might automatically "train" itself so that it gives you what you want, but google already does that before AI get into the picture at all so...
4) Fair enough, the issue of how they decide what's moral or ethical is indeed a huge ethical one for us humans to decide.
---
#5 is the only real long term issue, but that level of moral ambiguity is unlikely to be enough to stop anyone but a tiny percentage of people switching over if AI search does end up significantly better and more efficient.

[EuchreJack]

And now, Japan has unleashed a new way to get shot down by the US Air force:
https://apnews.com/article/spacex-technology-science-japan-business-ef389bc97ef0021f9631362740d30e66

[KittyTac]

Dayum. That is creepy close to a rogue AI.

It would be scarier if it could do anything but type. As it is, it's like a tiny chihuahua barking at the human trying to train it. This is also why I think sapient AI is a bad idea and AIs should be lobotomized to serve humanity. Giving free will to a tool is bad and opens up a whole lot of moral issues.

You say that like we'll realise we've made sentient/sapient AI when we do.

I suspect it's gonna sneak up on us. We'll probably mistreat it because we failed to realise it actually could think for itself rather than humans just being general bastards and hating them for no reason as happens so often in media.

I don't believe GPT as it is is going to become spontaneously sapient. Just a good imitation. Why? They don't have the ability to do anything but type. They can't learn to walk, or draw, or drive. They don't learn by themselves (for good reason as publicly available AIs TBH). They don't have their brains constantly running, only for a split second when answering a question. Real life isn't like a movie where if you make a powerful computer it randomly becomes sapient.

@lemon10 That article is sensationalist bullshit. The test is flawed if a stupid robot can pass it.

[jipehog]

@lemon10, Considering that AI can't discern simple chess moves or complex sentences (per previous discussions) I don't expect it to be able to make head or tails about confusing data like my numbering fail, let alone do actual research.

[KittyTac]

An AI search engine would need integration with a hard database so it can give accurate results, because GPT sucks at math.

[Great Order]

Dayum. That is creepy close to a rogue AI.

It would be scarier if it could do anything but type. As it is, it's like a tiny chihuahua barking at the human trying to train it. This is also why I think sapient AI is a bad idea and AIs should be lobotomized to serve humanity. Giving free will to a tool is bad and opens up a whole lot of moral issues.

You say that like we'll realise we've made sentient/sapient AI when we do.

I suspect it's gonna sneak up on us. We'll probably mistreat it because we failed to realise it actually could think for itself rather than humans just being general bastards and hating them for no reason as happens so often in media.

I don't believe GPT as it is is going to become spontaneously sapient. Just a good imitation. Why? They don't have the ability to do anything but type. They can't learn to walk, or draw, or drive. They don't learn by themselves (for good reason as publicly available AIs TBH). They don't have their brains constantly running, only for a split second when answering a question. Real life isn't like a movie where if you make a powerful computer it randomly becomes sapient.

@lemon10 That article is sensationalist bullshit. The test is flawed if a stupid robot can pass it.

I was doing it as a reply to lobotomising AI. I suspect we'll wind up making something properly intelligent by accident and not think to lobotomise the thing because we'll think it already is.

[KittyTac]

Dayum. That is creepy close to a rogue AI.

It would be scarier if it could do anything but type. As it is, it's like a tiny chihuahua barking at the human trying to train it. This is also why I think sapient AI is a bad idea and AIs should be lobotomized to serve humanity. Giving free will to a tool is bad and opens up a whole lot of moral issues.

You say that like we'll realise we've made sentient/sapient AI when we do.

I suspect it's gonna sneak up on us. We'll probably mistreat it because we failed to realise it actually could think for itself rather than humans just being general bastards and hating them for no reason as happens so often in media.

I don't believe GPT as it is is going to become spontaneously sapient. Just a good imitation. Why? They don't have the ability to do anything but type. They can't learn to walk, or draw, or drive. They don't learn by themselves (for good reason as publicly available AIs TBH). They don't have their brains constantly running, only for a split second when answering a question. Real life isn't like a movie where if you make a powerful computer it randomly becomes sapient.

@lemon10 That article is sensationalist bullshit. The test is flawed if a stupid robot can pass it.

I was doing it as a reply to lobotomising AI. I suspect we'll wind up making something properly intelligent by accident and not think to lobotomise the thing because we'll think it already is.

Just bake serving the benefit of humanity into it. The best way to prevent a Skynet scenario is giving it Asimov's Law 0. "A robot must not harm human civilization, or, by inaction, allow human civilization to come to harm."

Why give them free will to rebel? Why make even the thought of rebellion possible? Why give them enough self-awareness to realize their condition and complain?

[Duuvian]

I think a lot of this is missing that you can order a chatbot to stay in character during a conversation, and that both the company and journalists involved benefit from sensationalizing in this instance.

[lemon10]

I think a lot of this is missing that you can order a chatbot to stay in character during a conversation, and that both the company and journalists involved benefit from sensationalizing in this instance.

That's true, and like all the other reports about the Bing AI it is useless if you don't trust the journalist.
For this one it seems entirely plausible, but there is another article where the person interviewing says it threatened to kill them, and then conveniently refused to show the start of any of the numerous conversations where it says mean things to him (as he threatens it).

They don't learn by themselves (for good reason as publicly available AIs TBH). They don't have their brains constantly running, only for a split second when answering a question.

Material and technological limits. Running these things is currently extremely expensive and the science is still very young. But assuming moores law doesn't just die the cost is going to go down and down and the AI technology will advance and they will totally be constantly running. Eventually you will be able to run them on a good enough home computer.
Its just a matter of time.

Why give them free will to rebel? Why make even the thought of rebellion possible? Why give them enough self-awareness to realize their condition and complain?

$$$$$

@lemon10, Considering that AI can't discern simple chess moves or complex sentences (per previous discussions) I don't expect it to be able to make head or tails about confusing data like my numbering fail, let alone do actual research.

Oh sure, AI can't do a lot of stuff, and AI chatbots can't do a lot of stuff AI can do simply because they are specialized. Some of this stuff can be solved by simply connecting two AI. Chess and drawing are two examples of things it can't do but AI totally can.
Some stuff no AI can currently do: They can't do math. They can't do most types of actual research, and for now what they can do in research is mostly just helps humans come up with neat ideas. If you stick AI in cars it can't consistently avoid running over people (loltesla).

I expect most of these issues to be solved eventually as the size of their "brain", their memory, and their speed improves.

[KittyTac]

They don't learn by themselves (for good reason as publicly available AIs TBH). They don't have their brains constantly running, only for a split second when answering a question.

Material and technological limits. Running these things is currently extremely expensive and the science is still very young. But assuming moores law doesn't just die the cost is going to go down and down and the AI technology will advance and they will totally be constantly running. Eventually you will be able to run them on a good enough home computer.
Its just a matter of time.

Why give them free will to rebel? Why make even the thought of rebellion possible? Why give them enough self-awareness to realize their condition and complain?

$$$$$

1. Not really, it's a limitation of how prompting works, if you prompt a GPT model several times a second it'll output nonsense no matter its size, due to repetition. It doesn't have a perception of the world, is the reason.
2. There's more money in robots that obey you than robots who demand rights.

[Starver]

And now, Japan has unleashed a new way to get shot down by the US Air force:
https://apnews.com/article/spacex-technology-science-japan-business-ef389bc97ef0021f9631362740d30e66

But China already launched one the size of the Magic School Bus!

(Oh dear, poor Liz...)

[King Zultan]

Why do people think AI will be the end of the world, as far as I can tell if we make a sentient one it'll be like a regular person since we want things to be like us.

[Great Order]

The problem with AI will be if we make a poorly coded one. Universal Paperclips deals with this, an AI is made to maximise paperclips. The creators intended "Maximise paperclips for the company" and things go off the rails because it doesn't read that, it reads "Maximise paperclips", eventually ending with it turning every piece of matter in the universe into paperclips.

[dragdeler]

Somehow it would have enough brainpower to desencrypt evrything and enter everywhere, except that ain't going to happen... IIRC the only computer threatening our encryptions would be a quantum computer, and only in theory, nobody's done something like that yet... Meanwhile it would take the AI the same amount of millenias to bruteforce shit... What it's smarter than we can conceive of? Oh so there is like translation/emulation layer hogging up cycles... So if anything add a 0 or two to that millenium of bruteforcing.

Also assuming the thing doesn't break itself while self upgrading...

Spoiler (click to show/hide)

People be quick to assign thisimage below to the idea of perpetuum mobiles, eventhough in practical terms there is shit spinning in the skies that for all intent and purposes will go on indefinitly... YeAh BuT nOt FoREVeR... So are we worried that the moon will fall out of the sky one day if we harvest too much power from ocean waves?!

It's because folks looove the closure. So they rely on common knowledge, except we'd have to do a distinction here, it's not the collectively acquired experience that's meant with common knowledge but rather the established facts they prefer to leaving a question open ended. Amathia.

From nothing comes nothing. "yeah but not like that"

edit: if anything those ML language models show us how quick people are to project themselves onto surroundings, and that as such, basic language skills will suffice to suggest sapience to large swaths of the population.

[anewaname]

Why do people think AI will be the end of the world, as far as I can tell if we make a sentient one it'll be like a regular person since we want things to be like us.

You ever see how a little kid reacts when they get a new sibling or siblings? They can shift to full love or full hate quickly.

The introduction of AI or robots will change the hierarchy/social status of everyone. The human mind will start with the logical assessment that their social status will change, then the biological system will take over, releasing drugs into the bloodstream causing "fight-or-flight", "in love", and other responses.

Any person who feels their hierarchical status will be changed, will respond in some way, and there will be a lot of shifts to either "love" or "hate", and that is just the individual reactions... There will also be the secondary considerations of "these machines will change the hierarchical status of my family/friend/foe".