Using AI In Audio Debates & GR Research [Video]

[Hipster Doofus]

Amir 1, Danny Ritchie 0

JOKE ...so as always, it comes down to 1 and zeros

 

[Hipster Doofus]

Love your living room and equipment , that is, if it is yours of course

yes but the chair is facing the wrong direction

 

[beefkabob]

There is an AI policy here?

 

[amirm]

Yes: https://www.audiosciencereview.com/forum/index.php?threads/official-policy-on-use-of-ai.66323/

 

[BobbyTimmons]

Danny is the perfect person to start off with and then jump away from. He got me interested in audio science and components. It sounded right and showed discernable differences on charts and measurements. Then ASR came around, and there was a world of engineering knowledge to be had. Now the limus test, the anacdotal arguements. ASR's anacdotal arguements point back to measurements, electrical engineering, scientific method. GR Research points to anacdotal rules of thumb and emotion and scoffing at measurements others make while extoling his measurements. He then does the folksy thing, so that is surely trustworthy. It maybe comes to what tribe you want to belong to. I will say it seems much more fun and mystical to be with the Audiophiles. My pocketbook and ears like the Engineers and it turns out it kind of fun there too.

No, Danny isn't the "perfect person to start off with". He's the most idiotic scammer I've ever seen in the audio world. It's shocking anyone listens to him. Do they have IQs lower than 60?

 

[YSC]

AI thus far is something gimmicky and basically cyanide when dealing with focued, deep diving topic/issues as it basically just averages training materials like reddit posts...

 

[RexrothPigeon]

Death to clankers

 

[AudioX3]

Regarding using AI, I think you will find it very accurate regardless the subject if you do the right things. 1) Use the Pro or Thinking mode ONLY. 2) Verify what it says with further research if you don't know the subject well enough to understand is it halucinating, pleasing or taking BS and spitting it back. Fast mode is very subject to this as is Google search with AI, and same with Copilot and its use of Bing. 3) Ask a different AI engine the same thing and see if differences and that can lead to further assessing. 4) If you find yourself limited in time allocated for free of the Pro or Thinking mode, get a $20 subscription. You are paying for expertise in all subjects for a reasonable price.

In short, the fast mode needs to be treated as a toy or opinion all the way to right but highly suspect, especially that iit is relying on cached data and optimizing for speed, not up to date or correct. Pro and Thinking mode (Gemini and ChatGPT) are much better and reliable without any special prompts to get it to be sane and accurate.

Another aspect you will see as you start a AI chat journey, particularly Google AI / Gemini is if it has picked a path it wants to follow, no matter what prompt you give it, it will stray back to its false path. At that point just stop using it and go to another. That can happen with chatGPT as well in regular mode, but Thinking mode is significantly better in all aspects and is definitely net faster as well as you are getting much higher confident / competent info. I have subscription with ChatGPT and sticking with it.

Lastly, I have not tried Grok, and Claude is the one I would like to use, but I really dislike its response style of typing out it is like a telex machine or green screen. It is IMO a very bad marketing decision and very annoying and slow to spit things out to the point I just can't use it.

 

[antcollinet]

Regarding using AI, I think you will find it very accurate regardless the subject if you do the right things. 1) Use the Pro or Thinking mode ONLY. 2) Verify what it says with further research if you don't know the subject well enough to understand is it halucinating, pleasing or taking BS and spitting it back. Fast mode is very subject to this as is Google search with AI, and same with Copilot and its use of Bing. 3) Ask a different AI engine the same thing and see if differences and that can lead to further assessing. 4) If you find yourself limited in time allocated for free of the Pro or Thinking mode, get a $20 subscription. You are paying for expertise in all subjects for a reasonable price.

In short, the fast mode needs to be treated as a toy or opinion all the way to right but highly suspect, especially that iit is relying on cached data and optimizing for speed, not up to date or correct. Pro and Thinking mode (Gemini and ChatGPT) are much better and reliable without any special prompts to get it to be sane and accurate.

Another aspect you will see as you start a AI chat journey, particularly Google AI / Gemini is if it has picked a path it wants to follow, no matter what prompt you give it, it will stray back to its false path. At that point just stop using it and go to another. That can happen with chatGPT as well in regular mode, but Thinking mode is significantly better in all aspects and is definitely net faster as well as you are getting much higher confident / competent info. I have subscription with ChatGPT and sticking with it.

Lastly, I have not tried Grok, and Claude is the one I would like to use, but I really dislike its response style of typing out it is like a telex machine or green screen. It is IMO a very bad marketing decision and very annoying and slow to spit things out to the point I just can't use it.

All well and good, but I've had the paid thinking tiers come back with complete nonsense also.

It is not a machine you can just ask a question of and get a 100% reliable answer, no matter what tier you have. It is a machine you can work with in dialogue mode to solve problems. Treating it as a competent colleague - who doesn't necessarily have all the answers - works wonders. Let it suggest stuff - try it out - validate it - try again when that doesn't work. Eventually the correct result will emerge.

 

[AudioX3]

All well and good, but I've had the paid thinking tiers come back with complete nonsense also.

It is not a machine you can just ask a question of and get a 100% reliable answer, no matter what tier you have. It is a machine you can work with in dialogue mode to solve problems. Treating it as a competent colleague - who doesn't necessarily have all the answers - works wonders. Let it suggest stuff - try it out - validate it - try again when that doesn't work. Eventually the correct result will emerge.

I think we on same page. The one thing i will say, and trying to help others with, it is a journey and after 8 mths I can give the group a fast forward to use the Pro and Thinking modes, when ready try a free 30 day subscription, and be wise when using AI. I think you may be saying the same thing regarding being wise, but I also can attest it is tremendously useful and accurate with the proper due dilligence verification depending upon the criticality of the question asked. I don't want those who have had bad experiences or not tried AI and get the impression it is very often coming back with garbage when it in Pro or Thinking mode, it is much better but not infallable.

For example, a legal based question, definitely double check it, but if you ask it how to accomplish a task in excel and it gives you a complex formula that you then plug in and it works, that just saved you possibly 4 hours or infinity if it beyond something you have the mental capacity to calculate.

The fear that is comes back with BS, halicinations and dangerous answers is real, but do not dismiss AI with those fears, use it and be wise about it . Don't use fast or standard mode, and you will have a much better experience and develop a methodology of when you can trust it or have pause. And if have pause, take the wise action to triangulate or research further. And lastly, I can pay a lawyer $1000/hr and get the wrong answer (or right one), and AI is not flawless either. It is still trying to be a human just with increadible powers. AI actually scares the s#$t out of me as to where it is going to be misused, but it can be amazingly useful until it one day destroys the earth or saves the earth.

A real example pertainent to ASR, I just had ChatGPT do analysis and pros/cons in picking the right Buckeye amplifier for my desired, situations and before that it helped me select down to Buckeye.

 

[Hapo]

...infinity if it beyond something you have the mental capacity to calculate.

 

[testp]

this is a little off topic, but i thought i'll mention it regarding fear of AI taking jobs. proponent AI ppl keep saying ouh it will just revolutionize current job market, and there's gone be other jobs openings coming (i've even heard some say, hey there's noting wrong with plumbing jobs etc), but the key issue i think is overlooked is to where all of us ppl are being channeled. current AI looks like is slowly taking high-level skilled jobs (something we really need to keep our brains challenged - it's like food for brain).

so whenever some AI proponent is saying something like, ouh there's noting wrong with plumbing job, - and i'm sure there isn't, but i would like ask them back: are you high or just stupid.

anyway, the AI cat is out of the bag, so no way to put that cat back in, but i really hope those new jobs will be as rewarding for our brains as the ones currently overtaken by AI.

 

[Somafunk]

anyway, the AI cat is out of the bag, so no way to put that cat back in, but i really hope those new jobs will be as rewarding for our brains as the ones currently overtaken by AI.

The most rewarding job I had was taking troubled kids, either one on one or groups of 4 max out into the hills/mountains and either teaching them mtb skills/bike repair or building camps/structures, teaching them navigational skills/how to read weather/plot routes etc.

I’ve never had a job that could in any way be replaced by a.i.of any kind

 

[RexrothPigeon]

I think you will find it very accurate regardless the subject if you do the right things. 1) Use the Pro or Thinking mode ONLY. 2) Verify what it says with further research

Am I misunderstanding or is #2 basically saying "It is accurate if you double-check to make sure it is accurate"? If it is "very accurate regardless of the subject," I shouldn't have to verify with further research. (Of course it is not very accurate and everyone using these toys should verify before taking seriously anything they barf up)

 

[AudioX3]

Am I misunderstanding or is #2 basically saying "It is accurate if you double-check to make sure it is accurate"? If it is "very accurate regardless of the subject," I shouldn't have to verify with further research. (Of course it is not very accurate and everyone using these toys should verify before taking seriously anything they barf up)

I strongly disgree with that in all aspects not just AI. Be it a lawyer, doctor, amir or anyone or anything else. It is not close to a toy unless you use it like a toy and hopefully then only for toy subjects. I understand however how you can be highly skeptical, but suggest you trying it in a manner I @antcollinet ] or @amirm has suggeted. Most important is to use Pro or Thinking mode for anything you deam other than social conversation. And if it a subject you don't know, get educated and verify especially if the subject is important. So I would change your "it is accurate if you double-check to make sure it is accurate" to what I am trying to convey which is it is highly accurate in the advanced modes, but still you want to verify in some manner. Verification could be as simple as you use the formula it gives you and see that it gives back known correct answers. You ask for its sources and reference points. If it is an IRS question and it points you to the exact IRS form and words, that a very valid check. If you want to use it for analysis of amplifiers, you want to double check against a trusted source like ASR / amir reviews. But it can get you to a great fast starting point saving a lot of time and pointing you in the right direction. You can also help it with prompts like I want a scientific analysis not opinions.

The danger with AI that I fear is in the hands of anyone like a manager or executive or the military or a politician who just executes whatever it says, if it is using bad data like a business may have AI looks into its databases like as if that data is of confidence (it may be but it may often not be), this will compound bad decisions and make them even faster. I by career was a programmer and IT Architect, my mind thinks naturally in the terms of how to provide clear specifications to not go off course and with that AI is very valuable.

However as @amirm pointed out with GR Research, he used standard mode, we have no idea what he asked, he abused what responses he liked and left out what he didn't. In short, AI just acted like Danny Richie. If he used Pro or Thinking mode, it very well likely would have come back with scientifically sound answers, not most frequently found answers or the desire to please and echo. That still doesn't mean Danny would not cherry pick or abuse, it just would make it harder or make him switch back to fast mode.

Lastly, if you want to be AI free, that fine too. That is really not a bad choice at all.

 

[RexrothPigeon]

So I would change your "it is accurate if you double-check to make sure it is accurate" to what I am trying to convey which is it is highly accurate in the advanced modes, but still you want to verify in some manner.

I think we have different understandings of "highly accurate." To me, a calculator is "highly accurate." A well calibrated thermometer or barometer is highly accurate. Klippel NFS is highly accurate.
I use LLMs all the time. They're useful. But they remain toys: sometimes they can be used as tools (I can make a plate or something out of Legos; it will work) but since--as you say--you have to verify, they're too risky for anything mission-critical since they cannot be trusted (I wouldn't send Lego tanks into war).

 

[AudioX3]

No, Danny isn't the "perfect person to start off with". He's the most idiotic scammer I've ever seen in the audio world. It's shocking anyone listens to him. Do they have IQs lower than 60?

Perfect person for me to have found is a better way to put it. Just like Bose 901s. It got me enthused, started down a direction, but then a realization that there is better, much better. Danny however along with New Record Day got me into open baffle speakers and that led me to my Spatial Auido Labs M4s which I absolutely love. But I sure glad I didn't go down the GR Research kit mode rather than buy from an honest player like Clayron Shaw who is also friendly with New Record Day. But you are right to call out the "Perfect person" but it was meant to be a transitional step.

 

[AudioX3]

I think we have different understandings of "highly accurate." To me, a calculator is "highly accurate." A well calibrated thermometer or barometer is highly accurate. Klippel NFS is highly accurate.
I use LLMs all the time. They're useful. But they remain toys: sometimes they can be used as tools (I can make a plate or something out of Legos; it will work) but since--as you say--you have to verify, they're too risky for anything mission-critical since they cannot be trusted (I wouldn't send Lego tanks into war).

I get your point. A better phrasing on my part may be "often accurate" rather than "highly accurate". It is a probability machine in the end. But your qualification of highly accurate being a machine to be able to count on as right, I agree it is not. I do think it is an overstatement to qualify it as a toy that often spits out garbage. It is a very useful tool in the right hands and the right subjects which indeed are broad. I have used AI to help me make right decisions I may not otherwise have made or attempted without it and that is broadly in terms of law, irs, audio, books i may like, education, finance, accounting, taxes...that is pretty darn powerful tool not a toy. And to me a lot of that was critical and extremely helpful in getting accurate answers to very important questions. 60% maybe it was right first stop. 15% it is damn good that i double checked and got it on track as it misinterpretted my request....And just like a "help desk" "expert" it has as much as 10% sent me completely in the wrong direction. That last sentence has greatly decrease with advance modes. I just trying to share, I am definitely not evangelizing it is anything close to a single function precision machine that you gave as example of highly accurate devices, and again in using your definition I 100% agree highly accurate should better be phrased as often accurate.

 

[Keith_W]

I think we have different understandings of "highly accurate." To me, a calculator is "highly accurate." A well calibrated thermometer or barometer is highly accurate. Klippel NFS is highly accurate.

That reminds me of this graphic:

 

[RexrothPigeon]

That reminds me of this graphic:

View attachment 534433

Yes, I'd put llms circa 2026 in the low accuracy/low precision quadrant. they simply can't be relied on without additional research/verification.