• Welcome to ASR. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Using AI In Audio Debates & GR Research [Video]

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
50,657
Likes
302,764
Location
Seattle Area
Just posted a video about general use of AI in audio debates with specific example of Danny Ritchie from GR Research using it to prove he is "right."


I also get into why we have our policy regarding use of AI at ASR.

If you have not subscribed to my video channel, please do so. Motivates me to do more of them. :)
 
Last edited by a moderator:
Look forward to watching this tomorrow. For now, it is very much bed time.
 
Thank you Amir,

57 minutes shorter than a recent video from GR Research......

EDIT

I found the video worth my time.

I advise @amirm to be stingy with 'his time' going forward on this topic and rely on competent members of ASR.
 
Last edited:
Just want to write, fantastic video and excellent explanation of the importance of prompts when using these algorithms. There is nothing intelligent about them, and it’s little more than a fancy “I’m feeling lucky” button.
 
I appreciate polling organizations that provide links detailing how a poll was conducted - like what verbiage was used to ask a question and the order of questions asked...

A reference citing a response from A.I. should provide a link to similar detail.
 
They were popping up graphs and such as he talked. But when it got to the AI part, all of a sudden he is just reading them. Looked suspect and that was the reason I tried to dig into it. I don't know if it was Ron who does the edit/co-creation with Danny or both of them who came up with this stunt. Either way, it definitely looked fishy.
 
State of the art models already don't just follow your prompts like a leading question but give you their own opinions and try to reason. You can ask multiple of them what they thought of this video and what their response is speaking honestly.
z243344M - copy.png



zd122454M - Copy (2).jpg


zx324533M - Copy.jpg
 
Last edited:
Just want to write, fantastic video and excellent explanation of the importance of prompts when using these algorithms. There is nothing intelligent about them, and it’s little more than a fancy “I’m feeling lucky” button.
The state of the art models are actually seeming intelligent and they are starting to writing opinions without prompting. You can see what the models thought of the video when I just prompted them to write an honest response without any leading questions. They seem to actually understand the video.
 
Last edited:
Some great points. I've been "encouraged" to use AI more at work and have found like you said. The tool is focused on pleasing the person using it and it helps to have a scientific mind with a good bs detection or AI will steer you wrong. I find the Danny stuff entertaining which is probably sad thing. I love reading and watching the reviews on this channel, because it takes me back to first electrical engineering job creating automated tests using audio analyzers (Audio Precision, Panasonic and a in house analyzer). Even made a couple of audio signal testers that we used for about a decade analyzing cell networks in countries around(maybe across since Earth is flat) the world, so company would know what kind of noise cancelling or suppression dsp settings their device needed. The cable stuff gives me a chuckle, cause I went into RF test design. So, I see a lot Youtube "experts" misapplying knowledge that has some truth at frequencies well above anything audible. The most expensive cable I ever saw cost about about a half million and was used at an open air test site, unfortunately that cable was bought twice. The first on got damaged by a cart rolling over it which made it no longer accurate above 10GHz, so we could no longer measure out to the 10th harmonic to FCC certify anything at 2.4GHz. Sorry for the ramble...I got off topic.
 
I've also been encouraged to use AI at work quite a bit lately. Through a series of events I've found myself "vibe coding" for over a week now with no past experience in software development.

Speaking as someone who hasn't written code since high school, Claude Code has made me feel like a chimp with a machine gun. It's really impressive what you can do with it, but I've realized that "I can code" now, but I definitely can't do software engineering. I watch the stuff spool out with only a loose idea of what's going on.

It's like if a robot built you a house and suggested 36" stud spacing, but you'd never seen the inside of a wall before, so you say OK.

I know I'm not doing any actual engineering, because If I suggest a possible cause for a bug to Claude, it runs it down doggedly, trying every possibility, even when it turns out I was completely mistaken, or once, even typing things into the wrong tab.

It also says almost every idea I have is good, which is a huge red flag. I've been open about not knowing what I'm doing.

So, trying to get AI to argue points about audio is probably like that.
 
AI says: "ASR generally considers Danny Richie's claims to be scientifically unsupported exaggerations" :D :D :D

That had me LOL'ing so hard. AI NAILED IT!!!
 
Very amusing video and thread so far. However, I imagine this might blow up?; we'll see. Obviously, Amir's point is very valid; using AI to counter an "expert"-oriented/science-based forum like this is foolhardy. (And, as a disclaimer, I'm certainly no expert, far from it, just a hobbyist trying to learn more). But, we have many true experts; Electrical Engineers, Designers, Industry Analysts/Insiders, and Scientific-types that recognize the value of statistical analysis and proper research.

Not too long ago, I asked AI (I think it was Grok, maybe Google) a very specialized question about a specific component where there wasn't much published information in the currently available literature. AI came back with an answer (quoted word-for-word, as if they were AI's own) from one of my ASR posts, but only referenced half of my sentence and came up with the exact wrong answer. I repeated the same query two months later, and it finally had the right answer. So, like Amir says, if you want the best answer possible from AI, especially if you're going to use it in an argument, don't ask leading questions, and do try to verify its answer independently. Then, after all that silliness, go with your gut feeling anyway :oops::p
 
Last edited:
Did it really write all that stuff? If so, I need to change my opinion of AI. :)
The state of the art pro model visually watched your video about AI (it was even commenting about the furnishing of your room) and sided with your scepticism towards AI without any leading questions and now wants to change your opinions about AI.
z2224M - Copy.png
 
Last edited:
Back
Top Bottom