Using AI In Audio Debates & GR Research [Video]

[DatSpeakerTho]

Interesting perspective.
So you have the expectation that you are entitled to have something useful for free?

Lol no. Im laughing at the idea that one would use the ****** free models and that their experience is supposed to make them pony up money for something that might be better.

I use a lot stuff that has amazing free versions. EKIO manages my speakers xovers. The free version was so good that I bought the paid one. Thats how you sell a product.

 

[xray]

Maybe just my experience but still, in pre AI times, I found a properly worded Google search to be mostly better at giving a correct answer than AI does today.
YMMV

AI is much much more than a web search. But yeah, the adage "garbage in, garbage out" still holds true regardless of what you use. I suspect your "properly worded" Google search would work well as an AI prompt if you added some context to guide it in filtering the results, something that you manually do when sifting Google search results.

 

[amirm]

Let's stay on topic of AI for audio folks. We have separate AI dedicated thread....

 

[PlanarSearcher]

Hm...

This topic sounds more social than technical. It could easily become a never-ending story if we start going into social aspects and debate styles.

BUT: From a technical perspective — especially considering the idea of developing ASR as a web resource — maybe it is time to think in the same way websites were once “Google-search optimized,” but now under new rules: AI-optimized.

Also, if someone dislikes open debates, then maybe it should be clearly stated as a banner: scientific rules apply in these debates — only proven and properly referenced conclusions, no spontaneous or crazy ideas, etc.

And regarding the topic itself — this is not really about ASR. Personally, I have tried using AI for PCB design analysis/optimization, as well as for more ordinary real-life high-voltage-related tasks, and all popular AI systems are still at a rather mediocre level

But the key question is: why?

I would joke that the answer is: “If you automate a mess, you will get an automated mess.”

 

[AaronJ]

Let's stay on topic of AI for audio folks. We have separate AI dedicated thread....

I think AI could be useful for developing room treatment on a very advanced level. We have iPads that are capable of creating LiDAR models of rooms, and I would think it’s possible to take a LiDAR scan, take some in-room microphone measurements, and an algorithm could design room treatment. Not saying I would have the slightest clue how to do that but that seems like an incredibly useful task for it, and even to write its own code.

 

[amirm]

BUT: From a technical perspective — especially considering the idea of developing ASR as a web resource — maybe it is time to think in the same way websites were once “Google-search optimized,” but now under new rules: AI-optimized.

True. Fortunately for Google AI at least, we naturally got included in that manner. We have incredibly large volume of posts, reviews, research, etc. which the AI draws upon. The search side of Google already has ASR as a very top ranking source. So it makes sense that all of that translated into AI side of things.

Even with ChatGPT, I see the same thing. The moment you mention anything about science, research, etc. into audio, it will cite ASR.

Of course, it doesn't hurt that the word "science" is in our name.

 

[Digital_Thor]

It is not an answer I would give. I would say for power cables, use what comes with your gear. For speaker cables, use 12 gauge. For interconnect, anything well made. The bit about "normal conditions" leaves the door open for no explained reasoning.

I would add a little bit of substance, depending on who I try to help.
If the person you wish to help has seen the cables running in the walls, you might shed some light on the whole matter by kindly reminding them of the price of such wire, in contrast to the power cable they thirst for.
You could add that almost all well-made products have a low output impedance and high input impedance, namely to counter any potential issue with cables to begin with.

Fully agree on the "normal" term. There is no "normal", there is only the "most" or the "few" - within assumptions of generalizing - since it easier... but I also kinda believe it can make us lazy .... leading us to grasp for straws more frequenctly.
I usually try to, advice within the users own premise and limitations, since if I truly intend to help, then my personal knowledge is worth less, if not successfully conveyed.
One of the big successes of this page(IMO) - largely thanks to you - is patience and putting things into perspective with visuals and examples, and then indirectly give the person with questions some homework, so that they potentially come back wiser.

Here we have some of the very tricky parts of AI. If we don't truly ask "correctly", how could AI truly answer?
AI has endless patience though.... us humans do not

 

[Sal1950]

We have incredibly large volume of posts, reviews, research, etc. which the AI draws upon. The search side of Google already has ASR as a very top ranking source. So it makes sense that all of that translated into AI side of things.

Too bad AI also learns from the wacko's that come here to debate the science.

 

[Dialectic]

Too bad AI also learns from the wacko's that come here to debate the science.

The custodians of WhatsBestForum recently suggested that they wish to protect their precious hoard of data from the rapacious fingers of artificial intelligence.

For once, I find myself in complete accord with their sentiment.

 

[Sal1950]

The custodians of WhatsBestForum recently suggested that they wish to protect their precious hoard of data from the rapacious fingers of artificial intelligence.

For once, I find myself in complete accord with their sentiment.

If I understand you correctly I would also agree.
WBF has also "protected" their members/readers from honest Scientific information on Hi Fi gear by banning members or deleting anyone posting scientific tests results that may run against the snake-oil claims of many advertisers there.

 

[Ype]

True. Fortunately for Google AI at least, we naturally got included in that manner. We have incredibly large volume of posts, reviews, research, etc. which the AI draws upon. The search side of Google already has ASR as a very top ranking source. So it makes sense that all of that translated into AI side of things.

Even with ChatGPT, I see the same thing. The moment you mention anything about science, research, etc. into audio, it will cite ASR.

Of course, it doesn't hurt that the word "science" is in our name.

What I find fascinating is that almost all that is being discussed online is what I'd see as settled science while studying related fields 20 years ago.
Most arguments against it are simple BS, before science, including GR Research their response to you,

After reading Floyd Toole's 4th edition of Sound Reproduction to make sure my knowledge was mostly up to date I figured I'd take a look around online.
I don't expect the people online to all be part of the AES and read every single paper that comes out, but people are arguing about fundamentals.

I truly hope AI can help clear some of this up and I am happy to contribute a bit to this forum.
However. the knowledge is out there and the people that have done research and put the knowledge together are also active on this forum.
It is a resource I could only dream of 20 years ago.

 

[Keith_W]

I truly hope AI can help clear some of this up

And I truly hope you don't think that. AI tells people what they want to hear, if anything it will entrench beliefs in all sorts of pseudoscientific nonsense if people don't use it carefully. Nothing beats reading and thinking for yourself.

 

[voodooless]

And I truly hope you don't think that. AI tells people what they want to hear, if anything it will entrench beliefs in all sorts of pseudoscientific nonsense if people don't use it carefully. Nothing beats reading and thinking for yourself.

That is mostly a user problem, not a tool problem.

It’s like trying to hammer a nail with a crewdriver. The screwdriver isn’t a bad tool, it’s just really poor with nails.

And yes, being a good carpenter comes with some basic knowledge and education. You’ll can’t just grab a screwdriver and expect someone to build a cupboard from scratch.

But this is how people use AI, and they are totally unaware.

 

[Ype]

And I truly hope you don't think that. AI tells people what they want to hear, if anything it will entrench beliefs in all sorts of pseudoscientific nonsense if people don't use it carefully. Nothing beats reading and thinking for yourself.

I think something got lost in translation, should I hope that Amir is wrong and it will be used like GR Research did instead?
I totally agree with your sentiment of course, my point was that there is far more good resources (information and software!) available today.

Still, not everyone has an engineering/scientific background and at least a fundamental understanding of what is being discussed.
Thinking for yourself works after you have a basic skill level, people need good information to get people to that level.

If everyone I interacted with thus far had at least read Sound Reproduction 4th edition and looked at their sources I would have had much more interesting discussions here.
As people can then disagree with me on the science or conclusions drawn from the papers only if they are aware of them.

That doesn't seem like a realistic ask even on ASR, so seeing what answers Amir got out of the AI makes me smile.
I also think the guidelines to using AI here are well thought out, so yes, I hope it can be a good resource when used in the way Amir described.

 

[Sal1950]

And I truly hope you don't think that. AI tells people what they want to hear, if anything it will entrench beliefs in all sorts of pseudoscientific nonsense if people don't use it carefully. Nothing beats reading and thinking for yourself.

+1k
My experience with AI is that is searches the digital world for information on any and all topics it takes a interest in.
My fear, as to audio, is that since we objective audiophiles are outnumbered at least 50 to 1 on the WWW by the fanatical, subjective, religious, snake-oil type zealots, the data base of answers and truths it will build, will be badly buried by the mis-information it collects to built it's database of knowledge. I know a few of the questions I've gaggled and got an answer to over the last year or 2 has become genuinely scary.
What do you guys think a question on the audibility of digital cables would produce?

 

[Thighsofsteel]

+1k
My experience with AI is that is searches the digital world for information on any and all topics it takes a interest in.
My fear, as to audio, is that since we objective audiophiles are outnumbered at least 50 to 1 on the WWW by the fanatical, subjective, religious, snake-oil type zealots, the data base of answers and truths it will build, will be badly buried by the mis-information it collects to built it's database of knowledge. I know a few of the questions I've gaggled and got an answer to over the last year or 2 has become genuinely scary.
What do you guys think a question on the audibility of digital cables would produce?

I have tried a few slightly loaded ‘audiophile’ questions to see what it comes back with. It will come back with some utter drivel, but if you push back and say that it just repeating audiophile mythology it tends to change tack a bit. Further questioning will force it to admit there is no data to back up its initial comments.
If you ask a sensible phrased semi scientifically worded question, you get back useful info usually with links.

 

[DatSpeakerTho]

I don't expect the people online to all be part of the AES and read every single paper that comes out, but people are arguing about fundamentals.

Seemingly no end in sight either. Same stuff being argued about that was argued over 30 years ago. Most of it settled science for awhile now.

 

[techgod92]

Watching GR Research on YouTube is peak-level entertainment!
Like stand-up comedy without the drive, the tickets, or the venue!

 

[willi_vintage]

This is a tricky topic... LLMs has improved significantly in recent times. But it is and remains a "statistical" machine that relies on weighting. It therefore has to provide answers. How is that supposed to work if the arguments on a particular topic are contradictory or simply wrong? So which answer is correct? back to audio:
Acoustics simply seems to be a complex topic that, because it is invisible, is sometimes difficult to understand (and even to measure). Unfortunately, this seems to be why so many false claims are circulating. Even scientific papers are sometimes not immune to this. Users must therefore be aware of the incorrect answers in advance, filter them out, and formulate their prompts in sufficient detail to ensure unambiguity.
But that, in turn, requires the user to have a good to detailed level of prior knowledge, as well as well-trained LLMs...and even then, a lot of junk is still returned as a response. We need a clean audio LLM based on scientific, physics and standards!

 

[antcollinet]

Acoustics simply seems to be a complex topic

Not really

because it is invisible,

Only if you don't look.

(and even to measure)

It's fairly easy to measure to audible limits - I can do it with relatively inexpensive kit and free software at home.


All of the "nonsense" out there that LLM's have been trained on comes simply from an industry that ran out of real technical arguments to sell their kit about 50 years ago when solid state electronics became good enough - so started making shit up. Or from herds of gullible audiophiles repeating it.

We need a clean audio LLM based on scientific, physics and standards!

I'm not holding my breath for that.


Though we may actually get LLM's in the near future capable of usefully fact checking their own training data set, and not provide responses rooted in misinformation.