• Security incident: ISF was recently accessed by intruders. Please change your password, and change it anywhere else you used it. Read more

Merged Artificial Intelligence

The Atheist said:
And even more than that off its peak, but still far more than it was 2 years ago.

They were market darlings for a short period.



I don't see that as a problem. Ellison has never been highly leveraged and the price would need to halve again before he has a personal problem.

We can but hope. Ellison is a disgusting human being.
Click to expand...
Oracle's whole future is in building data centres for companies that won't exist in a couple of years.
 
Ziggurat said:
No. They are able to structure words into forms that we interpret as ideas. LLMs do not have ideas in the way we have ideas. They do not actually think at all.
Click to expand...
Yes. Read Steenkh's post and what I see is an LLM pattern matching (very complicated pattern matching, I know a bit about tensors or used to) and identifying patterns that occurred in Rome that also occurred in Moscow and generates text. Very interesting. Potentially useful to the right person. Not intelligence.
 
Jack by the hedge said:
The future is Wargames meets Jurassic Park.

Run your AI agent in a sandbox, set it a task, and it tries to solve it by working out how to break out of its sandbox and then attack another AI company.

This is fine.
Click to expand...
Having worked for a major IT company then for a major customer thereof, I am wary of statements issued by any companies PR team.
 
Wudang said:
Yes. Read Steenkh's post and what I see is an LLM pattern matching (very complicated pattern matching, I know a bit about tensors or used to) and identifying patterns that occurred in Rome that also occurred in Moscow and generates text. Very interesting. Potentially useful to the right person. Not intelligence.
Click to expand...
No, sadly it's probably not even that complicated. You're overthinking it :P

As I have said before, since I know a bit of history, I can tell you that Russia point-blank claimed to be the the third Rome after the fall of Constantinople, because the current ruler had married a relative of the last Byzantine Emperor, so they figured out that they literally inherited the title through her. That's WHY the Russian Emperor was called Tsar ever after, which is a slight contraction of Tsesar, i.e., Caesar. (Same reason why the HRE Emperor got to be called Kaiser.)

But anyway, I digress. Recognising historical patterns has nothing to do with it. "Moscow is the third Rome" and the rationalisation for it are a claim that was made a lot, is repeated by historians, and even by some Russians with delusions of grandeur to this day. It's even verbatim in the Orthodox Wiki, as it's part of their church's claim to primacy. Or even in the regular Wikipedia. It's on Twitter and Telegram.

But anyway, I don't find it all that interesting at all that an AI managed to repeat verbatim a sentence from the Internet :P
 
Last edited:
Aye, that in any case. It's just that in this particular case it's the absolute most trivial thing. Whop de do, it managed to reproduce a short sentence from the Internet. The only way to make the bar any lower would be to be amazed at its using a word :P
 
Last edited:
Wudang said:
Yes. Read Steenkh's post and what I see is an LLM pattern matching (very complicated pattern matching, I know a bit about tensors or used to) and identifying patterns that occurred in Rome that also occurred in Moscow and generates text. Very interesting. Potentially useful to the right person. Not intelligence.
Click to expand...
The point seems to have been missed by most: there is no workable definition of intelligence (apart from "intelligence is what humans have"), so nobody can say if LLMs are intelligent. Very few humans come up with original thoughts more than a few times in their lives; most just rehash ideas they have become acquainted with earlier, sometimes in new and unexpected ways. In fact, pattern matching could be what the human brain does, because nobody actually know.

On this forum we have seen a novel written by an LLM that seemed completely original, in its plot, but possibly not in its writing style. Definitely something that would have been declared unique if written by a human, but is sure to be declared unoriginal when people know an LLM is behind it.

More than a year ago there was a primitive LLM playing some game, and researchers found that to their surprise, it had build up a representation of the map in its memory, even though as an LLM it was not supposed to be able to do that (I don't think it was particularly good at playing, but that is not the point).

I won't be devastated finding out that LLMs are not intelligent, but I certainly can't rule it out solely on the basis that it is based on pattern matching and a huge dataset.

I would respect contrary opinions more if they were based on a definition of intelligence that goes further than a requirement for intelligence to be based on "wet" material.
 
HansMustermann said:
No, sadly it's probably not even that complicated. You're overthinking it :P

As I have said before, since I know a bit of history, I can tell you that Russia point-blank claimed to be the the third Rome after the fall of Constantinople, because the current ruler had married a relative of the last Byzantine Emperor, so they figured out that they literally inherited the title through her. That's WHY the Russian Emperor was called Tsar ever after, which is a slight contraction of Tsesar, i.e., Caesar. (Same reason why the HRE Emperor got to be called Kaiser.)

But anyway, I digress. Recognising historical patterns has nothing to do with it. "Moscow is the third Rome" and the rationalisation for it are a claim that was made a lot, is repeated by historians, and even by some Russians with delusions of grandeur to this day. It's even verbatim in the Orthodox Wiki, as it's part of their church's claim to primacy. Or even in the regular Wikipedia. It's on Twitter and Telegram.

But anyway, I don't find it all that interesting at all that an AI managed to repeat verbatim a sentence from the Internet :P
Click to expand...
When a historian writes a book, are you just as dismissive when concepts are treated that have already been described in other books? Or is it only when an LLM discusses well-known concepts?
 
stanfr said:
I have not been following the AI threads...how much credibility is Roman Yompalskiy given? I think his actual quote was 99.9% doom within 100 years, but still...this reminds me of the Ai warnings from Dune, or the movie Forbidden Planet...

We're Doomed
Click to expand...
The title of that video is "AI Scientist: 99.9% Chance Super Intelligence Wipes Us Out By 2030". It sounds like clickbait to me, but 2030 is a mere 4 years away now, so I guess we'll find out soon enough. Within 100 years most of us will be dead anyway. I know I will. As for his credibility, I don't know. I only know that predicting the future is hard and a lot of very smart people have been wrong before when it comes to bold predictions of doom like this.
 
2030 seems a bit wild. While AI capabilities will be superhuman in most aspects, and social disruption will be way worse than today, I don't think we will delegate enough power to AIs to do some real damage.
But maybe the social disruption will be enough. We might simple start a war with China because of AI race, not because AI told us to do it.
Still my prognosis is: AI will get us in a way we didn't think about.
 
Last edited:
Puppycow said:
The title of that video is "AI Scientist: 99.9% Chance Super Intelligence Wipes Us Out By 2030". It sounds like clickbait to me, but 2030 is a mere 4 years away now, so I guess we'll find out soon enough. Within 100 years most of us will be dead anyway. I know I will. As for his credibility, I don't know. I only know that predicting the future is hard and a lot of very smart people have been wrong before when it comes to bold predictions of doom like this.
Click to expand...
A lot of 'smart' people predict doom to get fame and fortune. Compare that to the Exxon scientists who in 1979 predicted that fossil fuel consumption would cause 'dramatic environmental effects before the year 2050' and accurately calculated both the global CO2 level and temperature rise. No prediction of 'doom', just uncomfortable facts.

The uncomfortable fact now dawning on AI developers is that brute forcing it won't produce intelligence - it needs to have specific structures designed for it. Practical AI's of the future will typically have much smaller LLMs and more compute allocated to logic and reasoning. They will also have a 'conscious' supervisor constantly checking and guiding the output towards the goals we give them. They will be more moral and less less dangerous than humans because this 'conscience' will be hardwired into them.
 
Being unable to predict the future is fair, but it doesn't justify arguing against what IS the status now. Like that LLMs are actually reasoning. Maybe some future models will be able to. The current idea of a LLM still can't.

I mean, if you want an illustration, it's like the difference between someone claiming that cars will never fly, and someone taking that as reason to deny that car's currently don't fly.

It's AT BEST a textbook argument from ignorance, except even more nonsensical. A classical argument from ignorance is more like, "I don't know that pigs can't fly, so they can fly." While this one is more like, "I don't know if pigs won't ever be able to fly in the next million years, so that means they can fly." :P
 
Last edited:
You must log in or register to reply here.

Share this page

ISF - Join now!

Every member here is approved by hand. No bots, no spam, just people who care about evidence and honest debate.

Membership is free!

Create your free account

Back
Top Bottom