AI - Cheating or just evolution?

This forum is for anything not Reason related, if you just want to talk about other stuff. Please keep it friendly!
User avatar
crimsonwarlock
Posts: 2441
Joined: 06 Nov 2021
Location: Close to the Edge

Post 07 May 2023

avasopht wrote:
07 May 2023
An us-vs-them mentality.
That's mainly on "them", as the Deep Learning school have commenced in a smear campaign against every AI-technology that is not based on ANNs, for years now. Obviously to divert funding to their own projects. And it worked; it is near impossible to get funding outside of ANN/DL focussed projects. I deal with that on a daily basis.
avasopht wrote:
07 May 2023
But creating a substrate to cultivate the desired emergent behaviours and abilities is also a viable strategy.
Totally agree with that. However, there should be a scientific and/or technological explanation of that strategy being viable. Outside of that, it is just religion. ANNs, and their derivative technologies like Deep Learning and Transformers, have not shown any indication, scientific or technological, to be able to "emerge" from their initial training. People think that GPT is constantly learning, but it isn't. It is not how the technology provable works.

This also shows where things become hype: we see people saying things like "the next version is going to be so much better", while the simple question to that statement is: WHY? The mistakes that GPT4 makes were also in GPT3 and even in GPT2 before that. When looking forward from GPT2, it simply didn't get better. It is still making the same mistakes. You can only make these models larger, you can't make them different. And GPT2 -> GPT3 -> GPT4 has already proven that making them larger doesn't solve these problems. It is logical sense that even larger models will still not solve these problems.
avasopht wrote:
07 May 2023
Is that what he actually meant? Was he saying "ChatGPT makes mistakes and humans make mistakes, therefore they are equivalent"?
That is obviously what he was implying, at least that they are on route to something equivalent, as he uses all kinds of similar wording in every interview he has done so far.
avasopht wrote:
07 May 2023
Or was he saying that humans also make mistakes, therefore the mistakes should not be seen as a total failure?
I'm not convinced of that because the mistakes that GPT makes are not solvable with an ANN-based model. And he knows that, for sure. These mistakes ARE a total failure.
avasopht wrote:
07 May 2023
Obviously deep learning alone is not the way for AGI. Are any prominent experts in Deep Learning suggesting this (because it's a very naive idea)?
It is the fundamental premise that laid under the acquisition of Deep Mind by Google. Demis Hassabis has made this clear in several of his presentations, and that it is indeed his point of view. And he is not alone in that stance. Yann Lecun has held that mantra as well for years, and only recently diverged from that standpoint.
avasopht wrote:
07 May 2023
I tend to see positions eagerly dismissed as just "hype" and "fear".
I'm not saying that the positions are hype. But what's currently happening around it certainly is.
avasopht wrote:
07 May 2023
I don't think human beings are intelligent enough to create AGI by understanding general intelligence, and I think we're too dumb to realize that.
I don't agree with you on that, as it is my area of research (and dare I say, expertise). Again, there is quite a bit of understanding about what general intelligence is. You just won't find that understanding in the ANN/DL/LLM school of thought.
avasopht wrote:
07 May 2023
... but will take insight and inspiration from cognitive science, neurology and any relevant topics.
Exactly. And that is what is lacking in the ANN-focussed community. They still state that ANNs are the most "human-like" approach, while advances in neuroscience have shown this to be wildly untrue. The ANN school of thought has neglected other scientific areas, to the point that you can find research on this tunnel-vision behavior.
-------
Analog tape ⇒ ESQ1 sequencer board ⇒ Atari/Steinberg Pro24 ⇒ Atari/Cubase ⇒ Cakewalk Sonar ⇒ Orion Pro/Platinum ⇒ Reaper ⇒ Reason DAW.

avasopht
Competition Winner
Posts: 3979
Joined: 16 Jan 2015

Post 07 May 2023

crimsonwarlock wrote:
07 May 2023
And it worked; it is near impossible to get funding outside of ANN/DL focussed projects. I deal with that on a daily basis.
There's a strong "just let the DNN generalize it" cult. So it's even hard to make progress within machine learning.

During a presentation with a researcher who had successfully separated sound sources, I asked whether converting the frequency spectrum to a log2 scale had any benefits (which I would expect to be faster training and fewer layers). He said he had thought about it but that you can just let the neural network generalize.

Technically you can, but not with the architecture they were using.

--
crimsonwarlock wrote:
07 May 2023
Totally agree with that. However, there should be a scientific and/or technological explanation of that strategy being viable. Outside of that, it is just religion. ANNs, and their derivative technologies like Deep Learning and Transformers, have not shown any indication, scientific or technological, to be able to "emerge" from their initial training. People think that GPT is constantly learning, but it isn't. It is not how the technology provable works.
Hmm, ... I would expect this research to come from another department like engineering, physics or economics. But it is needed, especially for any objective to yield a particular type of emergent behaviour.
crimsonwarlock wrote:
07 May 2023
This also shows where things become hype: we see people saying things like "the next version is going to be so much better", while the simple question to that statement is: WHY? The mistakes that GPT4 makes were also in GPT3 ... It is logical sense that even larger models will still not solve these problems.
I think they are banking on being able to insert another model to solve that problem.

GPT has a pretty simple model. It's not that hard to tweak, and you might just find methods deriving from cognitive science being bolted on that way, or working alongside it.

They did this with ChatGPT, and it looks like Google is going the route of taking a more strategic route that doesn't rely on having such a large model.

Again, this suffers from the cult of "just let the NN generalize", so I'm a little on the fence here just because it's possible to be addressed.

I do think there is some provision with the right architectures, but there's not enough research (and its more CPU bound).
crimsonwarlock wrote:
07 May 2023
avasopht wrote:
07 May 2023
Obviously deep learning alone is not the way for AGI ...
It is the fundamental premise that laid under the acquisition of Deep Mind by Google ... And he is not alone in that stance. Yann Lecun has held that mantra as well for years ....
I think there will need to be a few more demonstrable breakthroughs for a real shift away from that thinking.

Or maybe just better tools for trying out new things. Right now, you need to have a very good idea of what you're doing to try anything new.
crimsonwarlock wrote:
07 May 2023
I don't agree with you on that, as it is my area of research (and dare I say, expertise). Again, there is quite a bit of understanding about what general intelligence is. You just won't find that understanding in the ANN/DL/LLM school of thought.
At least for me, I am considering the theoretical possibility of intelligence that transcends what we currently know, or just yields better results in a way that seems counterintuitive to us. We already see vast differences between even intelligent people. Some people just think in a way that allows them to run rings around others with a counterintuitive thought process.
crimsonwarlock wrote:
07 May 2023
avasopht wrote:
07 May 2023
... but will take insight and inspiration from cognitive science, neurology and any relevant topics.
Exactly. And that is what is lacking in the ANN-focussed community. They still state that ANNs are the most "human-like" approach, while advances in neuroscience have shown this to be wildly untrue. The ANN school of thought has neglected other scientific areas, to the point that you can find research on this tunnel-vision behavior.
Well, action potentials and how neurons interact bear absolutely no resemblance to ANNs, nor do they share any properties (beyond the potential for the accumulation of complexity through layers of interaction).

This is a huge shortcoming and it's a shame because there are so many areas of research that just won't advance.

There also needs to be more research on computational neurology to better understand how functionality compounds with neurons. It seems the only research being done on this front involves real neurons on a petri dish.

User avatar
bxbrkrz
Posts: 3864
Joined: 17 Jan 2015

Post 07 May 2023

Welp... I don't know about AI, but while reading this thread my brain has evolved significantly, expanding my mind beyond ether and reality. I will now light a candle and pray that AI be a good boy.
757365206C6F67696320746F207365656B20616E73776572732075736520726561736F6E20746F2066696E6420776973646F6D20676574206F7574206F6620796F757220636F6D666F7274207A6F6E65206F7220796F757220696E737069726174696F6E2077696C6C206372797374616C6C697A6520666F7265766572

User avatar
bxbrkrz
Posts: 3864
Joined: 17 Jan 2015

Post 07 May 2023

While we wait patiently to know if AI has a soul...


REPORT: Hollywood Studios Are Trying to Claim the Rights to Actors' Voices
Ahead of new contract negotiations, the actors union SAG-AFTRA reveals that some studios have attempted to claim ownership over performers' voices.

Image

A new report suggests some of Hollywood's major studios are taking a page from the playbook of The Little Mermaid's Sea Witch by seeking ownership of actors' voices.
Per The New York Times, the actors union known as SAG-AFTRA claims that its members have reported contracts which would give studios the right to use the signatory's voice to craft new performances in the future. The NY Times highlights a Netflix contract that would allow the company to simulate an actor's voice "by all technologies and processes now known or hereafter developed throughout the universe and in perpetuity."

The advancement of artificial intelligence software has made it possible for studios to simulate actors' voices with increasing accuracy. In partnership with Lucasfilm, a Ukranian company called Respeecher famously used archival recordings and an A.I. program to recreate James Earl Jones' Darth Vader voice, which Jones has aged out of, for Disney+'s Obi-Wan Kenobi series. Respeecher also used this technique to recreate the voice of a young Luke Skywalker (Mark Hamill) in The Book of Boba Fett. In September 2022, it was reported that Jones had signed his voice rights over to Lucasfilm and Respeecher so that the companies could continue to use his signature baritone in future Star Wars projects, including those produced after the actor's passing.

https://archive.ph/cBUxE

Funny how planet Earth is after all the center of the universe...
757365206C6F67696320746F207365656B20616E73776572732075736520726561736F6E20746F2066696E6420776973646F6D20676574206F7574206F6620796F757220636F6D666F7274207A6F6E65206F7220796F757220696E737069726174696F6E2077696C6C206372797374616C6C697A6520666F7265766572

User avatar
plaamook
Posts: 2594
Joined: 22 Jan 2015
Location: Bajo del mar...

Post 08 May 2023

I guess if you sign an agreement like that you get what you bargain for.
I’d imagine that most of the people they hire are already famous and prob aren’t that short of cash.
I mean…I’d sign it if the price was right but my phone ain’t exactly ringing off the hook for animated voice acting.
Perpetual Reason 12 Beta Tester :reason:

You can check out my music here.
https://m.soundcloud.com/ericholmofficial
Or here.
https://www.youtube.com/channel/UC73uZZ ... 8jqUubzsQg

User avatar
plaamook
Posts: 2594
Joined: 22 Jan 2015
Location: Bajo del mar...

Post 08 May 2023

Also…a bit of a tangent, but as you mention it… if you stand on any spinning celestial body it will appear that everything is spinning around you. So I forgive humans for that little hiccup of thinking they were at the centre of it all.
Perpetual Reason 12 Beta Tester :reason:

You can check out my music here.
https://m.soundcloud.com/ericholmofficial
Or here.
https://www.youtube.com/channel/UC73uZZ ... 8jqUubzsQg

User avatar
bxbrkrz
Posts: 3864
Joined: 17 Jan 2015

Post 08 May 2023

plaamook wrote:
08 May 2023
Also…a bit of a tangent, but as you mention it… if you stand on any spinning celestial body it will appear that everything is spinning around you. So I forgive humans for that little hiccup of thinking they were at the centre of it all.
As long as we:
are alone in the universe
never create and build human colonies anywhere in the universe.
Otherwise The Disney Corporation may have to create the Armed Space Contractual Enforceability Agency...

Was AI H.R. Giger in charge of the design?

757365206C6F67696320746F207365656B20616E73776572732075736520726561736F6E20746F2066696E6420776973646F6D20676574206F7574206F6620796F757220636F6D666F7274207A6F6E65206F7220796F757220696E737069726174696F6E2077696C6C206372797374616C6C697A6520666F7265766572

User avatar
plaamook
Posts: 2594
Joined: 22 Jan 2015
Location: Bajo del mar...

Post 08 May 2023

Why? You could be at the centre and move outwards.
As for aliens…who knew, right? 😂
Perpetual Reason 12 Beta Tester :reason:

You can check out my music here.
https://m.soundcloud.com/ericholmofficial
Or here.
https://www.youtube.com/channel/UC73uZZ ... 8jqUubzsQg

User avatar
bxbrkrz
Posts: 3864
Joined: 17 Jan 2015

Post 08 May 2023

The space alien lawyers could have the same type of contracts too... It will not end well for us, our AI are too dumb to detect all their loopholes. But hey! May the survival of the fittest lawyers win. It is the ultimate rule of Evolution :thumbs_up:

757365206C6F67696320746F207365656B20616E73776572732075736520726561736F6E20746F2066696E6420776973646F6D20676574206F7574206F6620796F757220636F6D666F7274207A6F6E65206F7220796F757220696E737069726174696F6E2077696C6C206372797374616C6C697A6520666F7265766572

User avatar
FGL
Posts: 427
Joined: 23 Jan 2015

Post 09 May 2023

Question: Wich Tool uses AI to Fake Voices, like John Lennon etc. ? .I dont see a public Tool around to Fake Voices like this without KI. Text to any Voice along Notes, where?
Music:
My Songs on Youtube: https://www.youtube.com/@FGLucas-wj1eg.
My Songs on Soundcloud: https://soundcloud.com/frank-georg-lucas.

User avatar
Quarmat
Competition Winner
Posts: 467
Joined: 11 Feb 2021
Location: Europe

Post 09 May 2023

FGL wrote:
09 May 2023
Question: Wich Tool uses AI to Fake Voices, like John Lennon etc. ? .I dont see a public Tool around to Fake Voices like this without KI. Text to any Voice along Notes, where?
Try Uberduck

User avatar
FGL
Posts: 427
Joined: 23 Jan 2015

Post 09 May 2023

Okay, thanks. But did this not need a underlying tool, without Ki?
Music:
My Songs on Youtube: https://www.youtube.com/@FGLucas-wj1eg.
My Songs on Soundcloud: https://soundcloud.com/frank-georg-lucas.

User avatar
bxbrkrz
Posts: 3864
Joined: 17 Jan 2015

Post 09 May 2023

757365206C6F67696320746F207365656B20616E73776572732075736520726561736F6E20746F2066696E6420776973646F6D20676574206F7574206F6620796F757220636F6D666F7274207A6F6E65206F7220796F757220696E737069726174696F6E2077696C6C206372797374616C6C697A6520666F7265766572

User avatar
Aosta
Posts: 1075
Joined: 26 Jun 2017

Post 12 May 2023

So now you can train robots to do sports? :shock:
To train a robot to do anything really...this is going to be really interesting.

Tend the flame

User avatar
bxbrkrz
Posts: 3864
Joined: 17 Jan 2015

Post 19 May 2023

757365206C6F67696320746F207365656B20616E73776572732075736520726561736F6E20746F2066696E6420776973646F6D20676574206F7574206F6620796F757220636F6D666F7274207A6F6E65206F7220796F757220696E737069726174696F6E2077696C6C206372797374616C6C697A6520666F7265766572

User avatar
bxbrkrz
Posts: 3864
Joined: 17 Jan 2015

Post 20 May 2023

757365206C6F67696320746F207365656B20616E73776572732075736520726561736F6E20746F2066696E6420776973646F6D20676574206F7574206F6620796F757220636F6D666F7274207A6F6E65206F7220796F757220696E737069726174696F6E2077696C6C206372797374616C6C697A6520666F7265766572

User avatar
bxbrkrz
Posts: 3864
Joined: 17 Jan 2015

Post 24 May 2023




Image
757365206C6F67696320746F207365656B20616E73776572732075736520726561736F6E20746F2066696E6420776973646F6D20676574206F7574206F6620796F757220636F6D666F7274207A6F6E65206F7220796F757220696E737069726174696F6E2077696C6C206372797374616C6C697A6520666F7265766572

User avatar
plaamook
Posts: 2594
Joined: 22 Jan 2015
Location: Bajo del mar...

Post 24 May 2023

‘An incredible productivity benefit for everyone’
Amazing…
Hey chatgpt…how do we send you back to the netherworld?!😃
Perpetual Reason 12 Beta Tester :reason:

You can check out my music here.
https://m.soundcloud.com/ericholmofficial
Or here.
https://www.youtube.com/channel/UC73uZZ ... 8jqUubzsQg

User avatar
Superology
Posts: 215
Joined: 24 Nov 2022

Post 25 May 2023

This AI is pissing me off, hahaha. I think it's more cheating and always will be. Even people themselves don't know how to make great music I mean hits. It's more the matter of luck. And it's more spiritual then we are used to think. But scientists suck last time. They don't want to make world a better place. They want to become famous. Such a shame.

User avatar
bxbrkrz
Posts: 3864
Joined: 17 Jan 2015

Post 28 May 2023

Superology wrote:
25 May 2023
This AI is pissing me off, hahaha. I think it's more cheating and always will be. Even people themselves don't know how to make great music I mean hits. It's more the matter of luck. And it's more spiritual then we are used to think. But scientists suck last time. They don't want to make world a better place. They want to become famous. Such a shame.
Image

"History Repeating"
757365206C6F67696320746F207365656B20616E73776572732075736520726561736F6E20746F2066696E6420776973646F6D20676574206F7574206F6620796F757220636F6D666F7274207A6F6E65206F7220796F757220696E737069726174696F6E2077696C6C206372797374616C6C697A6520666F7265766572

User avatar
Aosta
Posts: 1075
Joined: 26 Jun 2017

Post 09 Jun 2023

Well Meta and the Zuckmeister has just thrown down the A.I gauntlet in the race for full music creation and it is open source :thumbup: .

This is getting pretty close...5 years time? Gonna be crazy times for music makers. :?

Tend the flame

avasopht
Competition Winner
Posts: 3979
Joined: 16 Jan 2015

Post 12 Jun 2023

Aosta wrote:
09 Jun 2023
Well Meta and the Zuckmeister has just thrown down the A.I gauntlet in the race for full music creation and it is open source :thumbup: .

This is getting pretty close...5 years time? Gonna be crazy times for music makers. :?

I guess this is at the very least, a source of FREE sounds, right?

User avatar
bxbrkrz
Posts: 3864
Joined: 17 Jan 2015

Post 15 Jun 2023



:D



Image

Life and death false positives.
757365206C6F67696320746F207365656B20616E73776572732075736520726561736F6E20746F2066696E6420776973646F6D20676574206F7574206F6620796F757220636F6D666F7274207A6F6E65206F7220796F757220696E737069726174696F6E2077696C6C206372797374616C6C697A6520666F7265766572

User avatar
Aosta
Posts: 1075
Joined: 26 Jun 2017

Post 20 Jun 2023

And things keep evolving...

Tend the flame

User avatar
bxbrkrz
Posts: 3864
Joined: 17 Jan 2015

Post 28 Jun 2023

757365206C6F67696320746F207365656B20616E73776572732075736520726561736F6E20746F2066696E6420776973646F6D20676574206F7574206F6620796F757220636F6D666F7274207A6F6E65206F7220796F757220696E737069726174696F6E2077696C6C206372797374616C6C697A6520666F7265766572

  • Information
  • Who is online

    Users browsing this forum: CommonCrawl [Bot] and 2 guests