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: 2467
Joined: 06 Nov 2021
Location: ##########

Post 03 Dec 2023

avasopht wrote:
03 Dec 2023
Reasoning doesn't seem to fix optical illusions,
Reasoning does fix optical illusions because we can reason with common-sense knowledge and decide that what we seem to see, is actually an illusion. Your statement here proves that point, when you point to "optical illusions" as something you can (obviously) reason about.
avasopht wrote:
03 Dec 2023
though yes, PRESENTLY we don't have introspective AI.
Actually, we do... but I can't tell you about it, or I have to kill you :puf_wink:
avasopht wrote:
03 Dec 2023
All I was getting at, however, is that we also make dumb errors of a similar degree so we should be cautious about being so smug about its mistakes
The problem I have with that, is that this argument is used as some sort of proof that current AI-stuff, as far as discussed in the media, is close to human-level intelligence. I think you agree with me that that is not the case.
-------
Reached the breaking-point. CrimsonWarlock has left the forum.

User avatar
bxbrkrz
Posts: 3962
Joined: 17 Jan 2015

Post 03 Dec 2023

crimsonwarlock wrote:
03 Dec 2023
avasopht wrote:
03 Dec 2023
Reasoning doesn't seem to fix optical illusions,
Reasoning does fix optical illusions because we can reason with common-sense knowledge and decide that what we seem to see, is actually an illusion. Your statement here proves that point, when you point to "optical illusions" as something you can (obviously) reason about.
avasopht wrote:
03 Dec 2023
though yes, PRESENTLY we don't have introspective AI.
Actually, we do... but I can't tell you about it, or I have to kill you :puf_wink:
avasopht wrote:
03 Dec 2023
All I was getting at, however, is that we also make dumb errors of a similar degree so we should be cautious about being so smug about its mistakes
The problem I have with that, is that this argument is used as some sort of proof that current AI-stuff, as far as discussed in the media, is close to human-level intelligence. I think you agree with me that that is not the case.
Was the human introspective coded?
757365206C6F67696320746F207365656B20616E73776572732075736520726561736F6E20746F2066696E6420776973646F6D20676574206F7574206F6620796F757220636F6D666F7274207A6F6E65206F7220796F757220696E737069726174696F6E2077696C6C206372797374616C6C697A6520666F7265766572

User avatar
crimsonwarlock
Posts: 2467
Joined: 06 Nov 2021
Location: ##########

Post 03 Dec 2023

bxbrkrz wrote:
03 Dec 2023
Was the human introspective coded?
Do you mean if human introspection is coded, I don't think so :lol:

Or do you mean if human-LIKE introspection has been coded... It definitely has, but as I said I can't go into that any further here :puf_bigsmile:
-------
Reached the breaking-point. CrimsonWarlock has left the forum.

avasopht
Competition Winner
Posts: 4058
Joined: 16 Jan 2015

Post 03 Dec 2023

crimsonwarlock wrote:
03 Dec 2023

The problem I have with that, is that this argument is used as some sort of proof that current AI-stuff, as far as discussed in the media, is close to human-level intelligence. I think you agree with me that that is not the case.
It's a brilliant illusion.

When you dig down what LLMs are doing and how they actually work and what their limitations are, it is an elaborate, though useful, trick.

Ditto for current neural network architecture.

It's amazing what it's possible and that it even works half as well, but its limitations speak volumes.

They're not really large language models, but statistical language model approximations with a baked in bias to produce statements in line with source data and an attention mechanism that allows prompting to, according to Andrew Ng, do roughly the same job of fine tuning.

User avatar
bxbrkrz
Posts: 3962
Joined: 17 Jan 2015

Post 03 Dec 2023

crimsonwarlock wrote:
03 Dec 2023
bxbrkrz wrote:
03 Dec 2023
Was the human introspective coded?
Do you mean if human introspection is coded, I don't think so :lol:

Or do you mean if human-LIKE introspection has been coded... It definitely has, but as I said I can't go into that any further here :puf_bigsmile:
For the first part of your answer :thumbs_up:
Is AI finally curious by 'nature", or does it still need a little push on that first domino?
757365206C6F67696320746F207365656B20616E73776572732075736520726561736F6E20746F2066696E6420776973646F6D20676574206F7574206F6620796F757220636F6D666F7274207A6F6E65206F7220796F757220696E737069726174696F6E2077696C6C206372797374616C6C697A6520666F7265766572

User avatar
selig
RE Developer
Posts: 12072
Joined: 15 Jan 2015
Location: The NorthWoods, CT, USA

Post 04 Dec 2023

bxbrkrz wrote:
03 Dec 2023
crimsonwarlock wrote:
03 Dec 2023


Do you mean if human introspection is coded, I don't think so :lol:

Or do you mean if human-LIKE introspection has been coded... It definitely has, but as I said I can't go into that any further here :puf_bigsmile:
For the first part of your answer :thumbs_up:
Is AI finally curious by 'nature", or does it still need a little push on that first domino?
My personal guess would be that we are still at the stage where AI is using only what is fed/taught to it. So in that case it would still need a push.
I understand some AI models have already “learned” (or created?) stuff on their own, like the two Facebook AIs that developed their own coded language to talk to (and negotiate with) each other. Granted, they were tasked with finding a way to negotiate IIRC:
https://www.independent.co.uk/life-styl ... 69706.html
So I’m wondering how close we are to AIs creating their own form of introspection or curiosity whether it has been coded/learned or not?
Selig Audio, LLC

avasopht
Competition Winner
Posts: 4058
Joined: 16 Jan 2015

Post 04 Dec 2023

selig wrote:
04 Dec 2023
bxbrkrz wrote:
03 Dec 2023

For the first part of your answer :thumbs_up:
Is AI finally curious by 'nature", or does it still need a little push on that first domino?
My personal guess would be that we are still at the stage where AI is using only what is fed/taught to it. So in that case it would still need a push.
I understand some AI models have already “learned” (or created?) stuff on their own, like the two Facebook AIs that developed their own coded language to talk to (and negotiate with) each other. Granted, they were tasked with finding a way to negotiate IIRC:
https://www.independent.co.uk/life-styl ... 69706.html
So I’m wondering how close we are to AIs creating their own form of introspection or curiosity whether it has been coded/learned or not?
The problem with AI (or anything) coming up with stuff on its own is that we humans don't entirely come up with stuff on our own. We exist in an environment. And when we come up with new ideas on our own, we can test them on the environment and others.

Sidenote: check out what is possible with things like LangChain (not sure if this paper used LangChain, but this is the sort of thing it was built for).


Using AI methods that don't (but can) involve neural networks, I've coded a Bomberman AI that does some unexpected things to win. These same methods when combined with neural networks changed our understanding of chess and Go because the AI found novel winning strategies that were counterintuitive to humans but made sense once discovered. But these algorithms are designed specifically to produce AI that comes up with its own plans (it's called Statistical Forward Planning).

These AIs, by the way, only ever played with themselves to get better.

And at the same time, we have predispositions to aesthetics that aren't entirely straightforward. What looks like a scary monster to most may appear "cute" to a psychopath.

On the other hand, human aesthetics could result from some deeper intrinsic beauty that would be appreciated by any intelligent being/entity/AI. After all, geometric shapes have notable properties.

Maybe aliens have also created something like dubstep, and all of our genres are natural conclusions to harmony, arrangement, rhythm and melody.

If that is the case, then the creative abilities of AI can be tested in greater isolation.

But without any frame of reference, a hypothetical creative AI without any human input could be hard for us to distinguish from random noise or complete garbage. Which is a risk human creatives face if they try to create works without any prior influences.
Last edited by avasopht on 04 Dec 2023, edited 1 time in total.

avasopht
Competition Winner
Posts: 4058
Joined: 16 Jan 2015

Post 04 Dec 2023

bxbrkrz wrote:
03 Dec 2023
crimsonwarlock wrote:
03 Dec 2023


Do you mean if human introspection is coded, I don't think so :lol:

Or do you mean if human-LIKE introspection has been coded... It definitely has, but as I said I can't go into that any further here :puf_bigsmile:
For the first part of your answer :thumbs_up:
Is AI finally curious by 'nature", or does it still need a little push on that first domino?
There's no "nature" to AI.

But Steve Grand's Creatures game did a good with inquisitive and creative AI. He writes about it in his book, Creation: Life and how to Make it.

User avatar
crimsonwarlock
Posts: 2467
Joined: 06 Nov 2021
Location: ##########

Post 04 Dec 2023

bxbrkrz wrote:
03 Dec 2023
Is AI finally curious by 'nature", or does it still need a little push on that first domino?
Curiousity is based on having intention. That is something the currently discussed models simply don't have, and why we shouldn't woryy about these things taking over the world :lol:

However...
selig wrote:
04 Dec 2023
My personal guess would be that we are still at the stage where AI is using only what is fed/taught to it. So in that case it would still need a push.
That is absolutely true for the Large Language Models and such that is being talked about in the media. But there are AI-projects that are not being talked about in the media :puf_wink:
-------
Reached the breaking-point. CrimsonWarlock has left the forum.

User avatar
crimsonwarlock
Posts: 2467
Joined: 06 Nov 2021
Location: ##########

Post 04 Dec 2023

avasopht wrote:
04 Dec 2023
These same methods when combined with neural networks changed our understanding of chess and Go because the AI found novel winning strategies that were counterintuitive to humans but made sense once discovered.
And then these novel winning strategies where completely crushed by humans without any prior experience in playing Go :lol:

It's explained in the first minutes of this video:


-------
Reached the breaking-point. CrimsonWarlock has left the forum.

avasopht
Competition Winner
Posts: 4058
Joined: 16 Jan 2015

Post 04 Dec 2023

crimsonwarlock wrote:
04 Dec 2023
avasopht wrote:
04 Dec 2023
These same methods when combined with neural networks changed our understanding of chess and Go because the AI found novel winning strategies that were counterintuitive to humans but made sense once discovered.
And then these novel winning strategies where completely crushed by humans without any prior experience in playing Go :lol:

It's explained in the first minutes of this video:


Yes, this goes back to what you said about us being able to reason about our decisions.

Although, it's not a real human without any prior experience - it was the researcher who wrote the algorithm to analyze the neural network weights to find weaknesses, then train an AI to exploit that weakness, observe the strategy, and play accordingly.

Still, I'd love to see what Lee Sedol thinks of this :)

User avatar
bxbrkrz
Posts: 3962
Joined: 17 Jan 2015

Post 04 Dec 2023

avasopht wrote:
04 Dec 2023
bxbrkrz wrote:
03 Dec 2023

For the first part of your answer :thumbs_up:
Is AI finally curious by 'nature", or does it still need a little push on that first domino?
There's no "nature" to AI.

But Steve Grand's Creatures game did a good with inquisitive and creative AI. He writes about it in his book, Creation: Life and how to Make it.
crimsonwarlock was correct: "by nature" meant intention in my context, but wasn't sure how to phrase that.
757365206C6F67696320746F207365656B20616E73776572732075736520726561736F6E20746F2066696E6420776973646F6D20676574206F7574206F6620796F757220636F6D666F7274207A6F6E65206F7220796F757220696E737069726174696F6E2077696C6C206372797374616C6C697A6520666F7265766572

User avatar
bxbrkrz
Posts: 3962
Joined: 17 Jan 2015

Post 04 Dec 2023

crimsonwarlock wrote:
04 Dec 2023
bxbrkrz wrote:
03 Dec 2023
Is AI finally curious by 'nature", or does it still need a little push on that first domino?
Curiousity is based on having intention. That is something the currently discussed models simply don't have, and why we shouldn't woryy about these things taking over the world :lol:

However...
selig wrote:
04 Dec 2023
My personal guess would be that we are still at the stage where AI is using only what is fed/taught to it. So in that case it would still need a push.
That is absolutely true for the Large Language Models and such that is being talked about in the media. But there are AI-projects that are not being talked about in the media :puf_wink:
Wink once if (you) are in the process of taking over the music business industry, or part of it.
Wink twice if neither confirm nor deny...
757365206C6F67696320746F207365656B20616E73776572732075736520726561736F6E20746F2066696E6420776973646F6D20676574206F7574206F6620796F757220636F6D666F7274207A6F6E65206F7220796F757220696E737069726174696F6E2077696C6C206372797374616C6C697A6520666F7265766572

User avatar
crimsonwarlock
Posts: 2467
Joined: 06 Nov 2021
Location: ##########

Post 04 Dec 2023

bxbrkrz wrote:
04 Dec 2023
Wink once if (you) are in the process of taking over the music business industry, or part of it.
Wink twice if neither confirm nor deny...
I can say this: it has nothing to do with the music industry. However, no telling where this might go, so it might end up there as well :puf_bigsmile:
-------
Reached the breaking-point. CrimsonWarlock has left the forum.

User avatar
bxbrkrz
Posts: 3962
Joined: 17 Jan 2015

Post 04 Dec 2023

crimsonwarlock wrote:
04 Dec 2023
bxbrkrz wrote:
04 Dec 2023
Wink once if (you) are in the process of taking over the music business industry, or part of it.
Wink twice if neither confirm nor deny...
I can say this: it has nothing to do with the music industry. However, no telling where this might go, so it might end up there as well :puf_bigsmile:
I believe it's a good thing to have a plaisant and amicable memories with all my future AI Overlords now, so they remember when they randomly access them.
757365206C6F67696320746F207365656B20616E73776572732075736520726561736F6E20746F2066696E6420776973646F6D20676574206F7574206F6620796F757220636F6D666F7274207A6F6E65206F7220796F757220696E737069726174696F6E2077696C6C206372797374616C6C697A6520666F7265766572

User avatar
bxbrkrz
Posts: 3962
Joined: 17 Jan 2015

Post 06 Dec 2023



Doxxed.....
757365206C6F67696320746F207365656B20616E73776572732075736520726561736F6E20746F2066696E6420776973646F6D20676574206F7574206F6620796F757220636F6D666F7274207A6F6E65206F7220796F757220696E737069726174696F6E2077696C6C206372797374616C6C697A6520666F7265766572

User avatar
Aosta
Posts: 1147
Joined: 26 Jun 2017

Post 11 Dec 2023

This would be so useful inside Reason, instant 'sampler' :eh:

Tend the flame

User avatar
bxbrkrz
Posts: 3962
Joined: 17 Jan 2015

Post 13 Dec 2023



Image
757365206C6F67696320746F207365656B20616E73776572732075736520726561736F6E20746F2066696E6420776973646F6D20676574206F7574206F6620796F757220636F6D666F7274207A6F6E65206F7220796F757220696E737069726174696F6E2077696C6C206372797374616C6C697A6520666F7265766572

User avatar
selig
RE Developer
Posts: 12072
Joined: 15 Jan 2015
Location: The NorthWoods, CT, USA

Post 14 Dec 2023

I’ve been wondering what is going to happen when chat AI starts eating itself, and we’re already beginning to see what looks like evidence of just that.
https://en.softonic.com/articles/grok-t ... ng-chatgpt

Is the next task of AI chat the ability to recognize other AI text in the wild to avoid ‘ingesting’ it?

One thing that keeps fascinating me is that chat AI is utterly dependent on the internet existing. Looking at the GIGO aspect (garbage in, garbage out), it seems there’s more work to be done on the training aspects of chat AI…
Selig Audio, LLC

User avatar
DaveyG
Posts: 2599
Joined: 03 May 2020

Post 14 Dec 2023

selig wrote:
14 Dec 2023
I’ve been wondering what is going to happen when chat AI starts eating itself, and we’re already beginning to see what looks like evidence of just that.
https://en.softonic.com/articles/grok-t ... ng-chatgpt

Is the next task of AI chat the ability to recognize other AI text in the wild to avoid ‘ingesting’ it?

One thing that keeps fascinating me is that chat AI is utterly dependent on the internet existing. Looking at the GIGO aspect (garbage in, garbage out), it seems there’s more work to be done on the training aspects of chat AI…
Yep, the first generation trained on the internet, which is created by humans, some of whom who are either wrong or who have specific agendas. The second generation trained on all that plus the musings and publications of the first generation. It can't end well.

Anyhow, I just asked ChatBotV3 to look into it and all it said was “I'm sorry, Dave. I'm afraid I can't do that”. Disappointing but at least it got my name right. :thumbup:

User avatar
bxbrkrz
Posts: 3962
Joined: 17 Jan 2015

Post 14 Dec 2023

selig wrote:
14 Dec 2023
I’ve been wondering what is going to happen when chat AI starts eating itself, and we’re already beginning to see what looks like evidence of just that.
https://en.softonic.com/articles/grok-t ... ng-chatgpt

Is the next task of AI chat the ability to recognize other AI text in the wild to avoid ‘ingesting’ it?

One thing that keeps fascinating me is that chat AI is utterly dependent on the internet existing. Looking at the GIGO aspect (garbage in, garbage out), it seems there’s more work to be done on the training aspects of chat AI…
Q. What is an authoritative source?
An authoritative source is a work known to be reliable because its authority or authenticity is widely recognized by experts in the field.
The Library specializes in collecting these types of resources so that students and faculty have the tools they need to research effectively. Rest assured that resources accessed through the Library’s website are always authoritative and appropriate for academic work.
Using Google (and other search engines on the Web) for research is risky. Free internet resources are sometimes authoritative (especially if they are offered by government agencies or academic institutions), but usually are not. Wikipedia, for example, is not authoritative because there is no way to verify authorship and anyone can edit an entry at any time.


https://piedmont.libanswers.com/faq/135714

1. Google Bart would disagree with their definition and position.
2. It is the same definition and position for each culture around the world. Each of them is the authoritative source.
3. Most if not all AI were trained on Wikipedia, the authoritative source.

We should be fine. No worries.
757365206C6F67696320746F207365656B20616E73776572732075736520726561736F6E20746F2066696E6420776973646F6D20676574206F7574206F6620796F757220636F6D666F7274207A6F6E65206F7220796F757220696E737069726174696F6E2077696C6C206372797374616C6C697A6520666F7265766572

User avatar
selig
RE Developer
Posts: 12072
Joined: 15 Jan 2015
Location: The NorthWoods, CT, USA

Post 14 Dec 2023

Next up, sound pack generation from text prompt.
Seriously!
Output is testing this in beta, I just got an email.
Stay tuned, let's see if Reason goes this route or similar.
Selig Audio, LLC

Higor
Posts: 123
Joined: 19 Jan 2015

Post 15 Dec 2023

selig wrote:
14 Dec 2023
let's see if Reason goes this route or similar.
I think this technology will become so commonplace that any online website will do the job. As said in a previous post, perhaps DAW's are a thing of the past.

User avatar
selig
RE Developer
Posts: 12072
Joined: 15 Jan 2015
Location: The NorthWoods, CT, USA

Post 15 Dec 2023

Higor wrote:
15 Dec 2023
selig wrote:
14 Dec 2023
let's see if Reason goes this route or similar.
I think this technology will become so commonplace that any online website will do the job. As said in a previous post, perhaps DAW's are a thing of the past.
We will still need a single place to gather all the audio files, even if it’s an online DAW or called something different by then.
As for Reason, they still need a way to create or browse patches for their synths - there are already WAY too many patches to ever keep track of by human means (especially without tags!). So even if external tools can generate custom audio, Reason still lacks a way to get the most from it’s vast library.
Imagine at least a “you liked this patch, maybe you’d also like this one” technology. Reason should ideally be your own personal studio “assistant” in as much as it’s possible, learning how you like to work and offering help to get to your goals faster. It won’t get you coffee, but it should be able to do most other typical assistant jobs. ;)
Selig Audio, LLC

Higor
Posts: 123
Joined: 19 Jan 2015

Post 15 Dec 2023

selig wrote:
14 Dec 2023
Next up, sound pack generation from text prompt.
i think results would be more easily achieved using reference tracks. But as an assistant text prompt might work well.

  • Information
  • Who is online

    Users browsing this forum: No registered users and 0 guests