I have asked so many audio questions I got banned from GROK. Wants me to spend $30 a month. Similar experience on ChatGTP. But Chat resets so many hours later. Able to resume questioning. However the program is not infallible. More than once I’ve accused the program of being the minister of misinformation. Telling the program I have lost confidence in the viability of the given information. The program answers sorry about that and I will try to do better. Overall a very positive experience. Sure a lot better than doing a DuckDuckGo search.
|
There is a lot of misinformation in AI-generated materials — I mean, a lot. Make sure you have the ability to judge and/or verify what’s right and what’s wrong. I see AI as a tool to help locate useful information, provided it's deemed reliable.
|
Out of curiosity, your post got me to ask ChatGPT to rate my system. It did a great job recognizing a lot of unique things about my system, but then it went it on to recognize pictures of my Dyna/VTA tube amps as Altec Lansing horn speakers!  . Not quite there yet!
|
If you think AI is getting scary close to being accurate with written answers, take a gander at what VEO3 by GoogleDeepMind can do. All the videos in this link are AI generated with just a prompt (using some basic parameters). From the video to the audio it's all AI generated, taking its cue from your prompt and taking it from there.
Soon, reality will be indistinguishable from fantasy run amok since we all know bad actors are going to capitalize on it.
All the best,
Nonoise
|
I built and trained a GPT on OpenAI's 4.o using Ivy League Master's courseware, as well as consulting back and forth about what it did and did not know, ultimately rolling the info back into the GPT's base training. I gave it my detailed system info and while it is many times better than when I started, it still struggles with some nuanced hallucinations, mostly resulting in it agreeing with me when it omits considerations (at least until I tell it what it did and to fix it).
I recently had a great conversation about my evaluation of a preamp replacement... moving from a decently modified Schiit Freya + to either a Benchmark LA4 or a Mark Levinson No. 5206. The objectives were to provide better line matching voltages and to keep the coloring of the buffering / follower stage of the sound to match the 1947 Telefunken AT12. It recommended the LA4 and also provided step by step instructions for how to match the Schiit Tyr input voltages.
FWIW - I have a separate phono preamp I am not looking to upgrade.
|
unfortunately chatbots are nowhere near reliable enough to trust answers to even trivial questions. you end up having to do a regular search to view the source material directly. and even when you correct it, the chatbot will eventually repeat the same errors. i was excited to use one at first, but rarely do any longer.
be warned, they are designed to stroke your ego and lure you into using them more and more. that is their real purpose, not providing accurate information.
|
I recently used both ChatGPT and Grok to help me narrow down which amp and speakers I should keep as I’m considering downsizing. I queried 4 amps and 4 pairs of speakers: Luxman LX-380, Luxman 550axMarkII, Audio Research i50, Musical Fidelity 6si and potentially a LTA Z40+. Speakers: Dynaudio Heritage Specials, Harbeth30.2XD, Devore Super Nines and Magnapan LRS. I added room dimensions etc, types of music I like, % streaming vs CD and vinyl and whether I listen when doing things around the house or stationary. I asked to list the pairing 1-2-3.
The results were quite similar between the two platforms and both gave very detailed summaries of their results. E.g. why the pairings make sense based on a number of factors. Anyone care to guess what I’m keeping?
|
I recently started using ChatGPT for comparisons. Sometimes the information is just wrong. When I pointed this out, it adjusted and provided different results. It is just like any other resource at this point. It is just retrieving information from the internet and re-packaging it, albeit a bit more efficiently than if I was pulling all the information together myself. I use it as a starting point. Ultimately, we need to determine what sounds best in our particular systems, unless you want to jump into the ASR pool and rely only on data points rather than what brings joy to you. I still trust my ears and judgement.
|
I use perplexity and asked it to analyze my system.
perplexity did a good job of providing a lot of detail about my system, and how it could be expected to perform. It did make a few suggestions, but they seemed pretty generic to me. It almost seems like a super search engine that rapidly assimilates what’s available on the net into a reasonably coherent package. This is obviously just a rudimentary AI tool, and I consider it useful, but not really intelligent, more of a compiler.
|
Yes, I've used it to ask a few audio questions and it gave me a list of equipment that some I had never heard of which was really cool. I think it works pretty good !
|
On ChatGPT there is a specific GPT named "Hifi Advisor" created by Borui Wang, that provides insightful views. However, I have also found that GPT bots often contradict themselves, especially when it comes to the deeper technical challenges, and on many occasions might exclude key parameters. (this then leads to me debating one GPT against another - quite entertaining).
For instance I was planning on building out a high-end HQplayer server on the new Mac Studio M4 Pro platform. After much deliberation with ChatGPT and specifications required for FLAC/DSD upsampling, PEQ & convulsion requirements on memory and CPU, I noticed elsewhere that the Mac USB does not support Native DSD. So when I asked ChatGPT on this limitation, it stated "oh, yes. that was a significant oversight on my behalf". So I wouldnt take everything as gospel, but it certainly provides guidance and thought-provoking considerations.
|
I have been using chatGBT, and found it to be outstanding and fun. While it makes some errors for example it thought my Moon 330A amp was NOT fully balanced, it is and thought my R11’s were 88 db sensitivity when they are 90. (still learning).
I have found it to be outstanding for tweeks. Which power cord to use on what component… component isolation strategies based upon component and shelving composition… system synergies etc.
It gives you a TOO rating:
(Total Overall Optimization), focused on sound-per-dollar impact, importance in the signal chain, and synergy with your current system, all while incorporating your room treatments and isolation efforts.
And perhaps most importantly, it never snipes. Outstanding
|
ChatGPT seems to good for simple sonic identity of components and tubes. But the more detailed questions I have posed that less I trust it to be right.
i have to remember it is compiling information it’s getting from the internet sources. Do I trust all I read on the internet? NO! Are review articles truth or opinion? AI is pulling info from them.
A simple example of inaccurate data as in a question regarding BAT VK-80i integrated amp it said the power tubes were KT88 when in reality they are 6C33C-B. For me that ruined its credibility.
That’s been my experience take it for what it’s worth, nothing
|
Another thing to consider. FEAR can mean false evidence appearing real. That’s how I look at AI.
|
ChatGPT can be a lot of fun wriing parodies of familiar songs.
Here's one I came up with this morning:
I asked ChatGPT:
"Write a parody song based on the theme of "Stairway to Heaven" about an elderly gentlemen who needs to desperately perform a "basic biological function" but discovers the first floor bathroom is out of order and has to "climb the stairway" to get to the second floor restroom."
ChatGPT said:
Sure! Here's a parody of "Stairway to Heaven" (by Led Zeppelin), called "Stairway to the Bathroom". It keeps the original's style—epic and dramatic—but with the comedic urgency of our elderly protagonist trying to find a functioning restroom.
🎵 "Stairway to the Bathroom"
(To the tune of “Stairway to Heaven”)
By Led Zeppeein-a-little
[Verse 1]
There’s a man who is old,
And his bladder’s quite bold,
But the bathroom downstairs is forbidden.
Though the sign says “Out of Use,”
He still rattles the noose,
As he clenches with growing misgivin’.
[Verse 2]
Ooh, and he groans as he turns,
Feeling internal burns,
With a shuffle that’s half-dance, half-pleadin’.
All the doors he passed by,
Offered hope, then denied—
He’s now cursing the plumbing and Eden.
[Chorus]
And he’s climbiiiiing… a stairway… to the bathroom.
[Verse 3]
His knee pops with a crack,
As he grasps at the banister,
Each step is a journey through time.
He's recalling a youth
Where each flush was the proof
Of a bladder that once was sublime.
[Bridge - spoken, whispery tone]
If there’s a bustle in your hedgerow…
That’s just him tryna hold it.
Don't be alarmed now—
It’s just a senior moment…
Yes, there are two stalls you can go to,
But in the end, there's only flushing one…
[Verse 4]
And as he reaches the top,
With his pants nearly dropped,
There’s a glow from the porcelain portal.
Though the light flickers dim,
It’s salvation to him,
Like a temple that's sacred… immortal.
[Final Chorus]
And he's fiiiiinally… arrived now… at the bathroom…
Where he’ll sit down… with triumph… in the stall-room…
Oh yes, he's maaaaade it… to the bathroom…
(He’ll be in there... ‘til next nooooon…)
Let me know if you want a version that’s more singable or you want chords added!
|
Unfortunately I see more than a few suspicious reviews. The kind that are pretty much playing with generalities like a clairvoyant searching your conversation for nuggets to build their faux understanding of their clients. Some of these are the vague "Barnum statements" of the Audiophile world so they can appear to understand you and to build a map and say pretty much what you want it to say.
|
For fun, I asked Chat GPT:
`Does the choice of power, interconnect, or speaker cable affect the sound a a high quality audio system?`
Answer:
Yes—but with significant caveats.
1. Power Cables
-
Effect: Minimal to negligible, assuming cables meet standard electrical specifications.
-
Why: Audio gear with well-designed power supplies already reject most line noise. As long as the power cable delivers sufficient voltage and current, there’s little room for improvement.
-
When it might matter: In environments with extreme electrical noise or poor grounding, upgraded cables with better shielding may reduce hum or RF interference. But it’s rare.
2. Interconnect Cables (e.g., RCA, XLR)
-
Effect: Small but can be audible—especially with poorly shielded or excessively long cables.
-
Why: Low-level analog signals are more vulnerable to interference and signal degradation. Balanced XLR cables reduce this risk.
-
What matters:
-
Placebo vs measurable: In controlled tests, audible differences are often subtle, if present.
3. Speaker Cables
Summary:
| Cable Type |
Audible Impact |
Conditions Where It Might Matter |
| Power Cable |
Low |
Poor power, high interference |
| Interconnect |
Low–Moderate |
Long runs, high RF/noise environments |
| Speaker Cable |
Moderate |
Long runs, low-impedance loads |
Final Thought:
Most audible differences attributed to cables in high-end systems are subtle and often subjective. Spending large sums on "boutique" cables yields diminishing returns—money is often better spent on room treatment, speaker positioning, or component upgrades.
|
I have used a Google product called NotebookLM. The advantage is that it will only look at the sources you provide. You can add youtube videos PDFs links. I think that this increases the quality of the answers.
I use a Google doc and paste in information that I find that is relevant. On a free account you can only add 30 sources, but the sources can be a large as you want. By adding content to only one Google Doc it reduces the # of sources..
https://notebooklm.google.com/
|
Then I asked:
`What do blind listening tests suggest about this topic. Restrict your comments to tests with high-end audio listeners as subjects':
Blind listening tests—particularly **double-blind** tests involving **experienced high-end audio listeners**—consistently show:
---
### 🎧 **1. Power Cables**
* **Finding**: No statistically significant audible differences under controlled conditions.
* **Example**: In tests conducted by audio engineering societies (AES), even audiophiles could not reliably distinguish between standard and high-end power cords.
* **Conclusion**: Perceived differences are typically due to expectation bias.
---
### 🎼 **2. Interconnect Cables**
* **Finding**:
* Differences are *occasionally* heard when cables are poorly made or excessively long.
* In most high-quality systems with properly made cables, **blind tests fail to confirm audible distinctions**, even among trained listeners.
* **Conclusion**: Brand or price does not correlate reliably with perceived quality when sighted bias is removed.
---
### 🔊 **3. Speaker Cables**
* **Finding**:
* Audible differences are more likely to be detected in blind tests **when comparing cables of significantly different gauge or impedance**, particularly with low-impedance or highly revealing speakers.
* Between two high-quality cables of similar spec, **even audiophiles rarely detect consistent differences blind**.
* **Conclusion**: The most important variables are physical: resistance, length, and termination—not brand or exotic materials.
---
### 📜 Representative Studies & Findings
* **Audio Engineering Society (AES) papers** (e.g., by David Clark, Sean Olive, Floyd Toole) often report that:
> Listeners—even experienced audiophiles—fail to distinguish expensive cables from inexpensive or standard ones in controlled, blind setups.
* **Boston Audio Society test**: A well-known blind test where listeners couldn’t reliably tell a Monster cable from a coat hanger wire in speaker cable comparisons.
---
### 🧠 Cognitive Bias Note
In blind tests, many effects attributed to cables disappear, suggesting that **sighted listening is heavily influenced by expectations, branding, and cost**. This is especially true in high-end audio communities, where psychological investment in gear is high.
---
### ✅ Summary for High-End Listeners
| Cable Type | Consistent Audible Difference in Blind Tests? | Key Variable (If Any) |
| -------------- | --------------------------------------------- | ---------------------------------- |
| Power Cables | ❌ No | None, if spec-compliant |
| Interconnects | ❌ Rare, unless defective or poorly shielded | Shielding, length |
| Speaker Cables | ⚠️ Occasionally, if mismatched impedance | Gauge, length, termination quality |
---
**Bottom Line**:
Even experienced listeners generally **cannot distinguish cables blind** unless a measurable electrical fault is present. Sighted evaluations often introduce psychological artifacts mistaken for sonic differences.
|
@cjmarkle - I agree 100%. These are tools designed to foster addictive behaviors. And addictive behaviors leads to tiered pricing. And tiered pricing leads to big income streams for these companies. I've worked in emerging technology on the west coast all my career. Although they like to paint their mission as virtuous, it's all about the money.
@maholl50 - My bet it suggested keeping the Audio Research I-50 and Devore's paired together. Based on your mentioned equipment, that is pairing I'd keep myself based on what I know about them. What did it suggest?
|
I've been using Google Gemini, standard issue on my Pixel 8. It can be very helpful but requires good prompts and follow up questions. I start by assigning an "expert" role related to my question. I'll tell it to be a "professor of electrical engineering", an "expert speaker designer", or similar. I also instruct it to seek out authoritative sources or specific sites. Then I follow up with requests and questions to resolve inconsistencies or hallucinations. This also prompts the AI to request more info from me to narrow down the information.
I don't use it for subjective evaluations of gear. That's my role. But I'll ask it to find reviews that mention a particular quality for suggestions.
|
As a casual user of AI. I have found its recommendations generally very good on sonic characteristics of equipment and general questions. But very bad at specific recommendations. It will often recommend the wrong piece of equipment... like I want a streamer and it recommends preamps or a mixture of different components.
Now if you are loading full sets of your own data and carefully crafting queries... then that is a different thing.
|
@dunham_john
You are definitely into some super sophisticated stuff. A very interesting comment.
to quote,
“provide better line matching voltages and to keep the coloring of the buffering / follower stage of the sound to match the 1947 Telefunken AT12.”
A 1947 Telefunken? I have a bunch from the early 1960s
|
Hey, @emergingsoul - Thanks. I have a typo in my comment. It's a 12AT7, not AT12.
As a follow-up, I previously went down the path of MELZ metal plate 6N8S tubes in both stages (before the adapters and tube changes to Telefunken and Sylvania) and the sound signature was a bit warmer, though not really improved. I also suspect with the mismatched output of the Freya+ and the inputs for the Tyrs (Why Schiit, why are they so far apart?), I might be able drive the Tyrs and downstream Magnepan 3.7is at higher volumes and lower distortion levels. My only concern is I'd be stripping what is happening with the Telefunken tubes and listening to my Tyrs alone. Anyone done this combo? At least Benchmark has a 30 day return policy if I buy it and it sounds awful.
|
It’s useful but you have to be careful with the data it returns. I’ve found it to be unreliable with certain details. For instance, I asked about a certain type of foam and it said it was closed cell but the foam was actually open cell.
|
I found it especially interesting in evaluating blood test results. Have to make sure I remain healthy to listen to music you know
|
i ID tubes for a lot of people, and the amount of disinfirmation coming out of chat GPT is already having ding dings think their idnorance trumps the accurate ID theyre given. ugh.
|
It certainly did “stroke my ego.”
|
Any wagers on how long until we live in this scenario?
All the best,
Nonoise
|
|
@waytoomuchstuff
If you need a road manager, I’m available for when you’re ready to take your band on the road.😁
|
|
I used ChatGPT and Grok to provide parametric EQs for my Roon using prompts like "provide me with roon parametric eq that will give me a warmer sound while maintaining 3d soundstage". Both AIs weren't entirely accurate but I had a lot of fun trying out what they suggested. As I am a complete parametric EQ beginner, the AIs provided a good learning experience as they explained what the frequency ranges do to the sound. Out of curiosity, I uploaded the frequency response of my speakers to them and the results are still not what I wanted. Ended up doing the parametric EQ myself with my learning. Will be fun doing the same exercise when the new versions come out.
|
@oliver_reid. When I asked chat gpt if cables made a difference to my setup I got very pro-cable feedback, unlike yourself. Apologies for the formatting, I copied and pasted from my phone. This is what it said:
1.Digital Cables (USB, Ethernet, SPDIF)
Many people assume “it’s just 1s and 0s”, but that’s a misunderstanding of how digital audio transmission works in real-world electronics.
What matters:
- Timing (jitter) — Especially important with USB and SPDIF. Poor cables can add jitter by causing reflection, signal degradation, or poor shielding.
- Noise injection (EMI/RFI) — Cables can act like antennas. In digital cables, power and ground noise can leak into DACs, affecting analog output.
In your USB-only setup (SOtM → Vitus):
- A better USB cable may improve:
- Imaging and depth
- Smoothness
- Low-level detail and microdynamics
- This is not because the bits are wrong, but because the timing and noise floor is improved.
✅ Many Vitus and SOtM users report clear differences with cables like Audioquest Diamond USB, Curious USB, or Shunyata Omega USB
2. Power Cables
Power cords can change the electrical noise entering your DAC or streamer and affect transformer vibration, both of which subtly influence sound.
What to expect:
- Cleaner bass
- Quieter background
- Slightly more tonal density
A quality shielded power cord with a tight ground reference can reduce EMI pickup and transformer hash. It makes more of a difference on DACs and streamers than, say, on power amps.
3.Interconnects (if you were using the Vitus as a CD player)
These matter most when passing analog signals, not digital. They affect:
- Capacitance (which alters HF extension)
- Ground interaction (can create hum/noise)
- Dielectric and shielding (influences tonality and clarity)
Since you’re using the Vitus as a DAC only, the analog interconnect between DAC and preamp or amp is still very important.
|
I received similar positive recommendations from ChatGBT on cables and power cords. It even went on to describe sound differences between my two upgraded PC’s & then recommend upgrading my DAC LPS PC. Interesting
it also focused a lot on my Sbooster. Upgraded power cords, weights, bamboo base, upgraded feet. Again interesting
|
Lots of the comments are reflective of marketing propaganda prepared by the manufacturers of the products. AI has no means to make a judgment about what it is saying.
I think most of us are Smart enough to discern meaningful comments amidst all the noise.
I don't like the ‘atta boy’ rhetoric when I ask a question. My ego does not need to be stroked by this device.
|
Re "AI has no means to make a judgment about what it is saying."
This is THE crucial point. The word "Artificial" should be kept in mind at all times.
|
That's why robots will not be good in the kitchen. Unless actions are clearly defined. Cooking requires continual reassessment and judgment. Also not very good taking care of the kids clearly. Although ability to think about what's being said clearly evolving quickly.
|
AI, is just a data mine. Right now, all it is doing is creating a data lake, using your inquires to gain more data.
Sometimes it gets it right, other times it does not. You still need to FACT CHECK AI.
|
|
@duckworp
Two completely conflicting sets of facts on the same question Who says AI can't learn, even on AG ;)
|
|
A word of caution, especially if you use the paid version of the tool. Since it remembers all you have said before, it often takes your subjective observations and doubts, for example experience during burn-in, as facts. And then it gives you an analysis based on your own subjective and changing considerations and other users' experience as if it were an objective comparison. Also I have an impression the tool kind of sides with you, disregarding objective factors, trying to please you...
|
By the way, if you want to use AI effectively, you should explore the subject of prompts you should ask before you begin. It is possible to train your AI to suit your purpose, but the matter of proper prompts is crucial.
|
Yeah, ChatGPT couldn’t figure out a McIntosh MC240 or MX110 tube schematic to save its life - 100% of the time (in my experience) mis identifying the tube and function for each tube. I’ve corrected it several times and it never remembers. That having been said, it’s an unbelievable tool that’s only getting better and when it comes to comparing synergy of systems and speakers or for separates it’s absolutely specific and almost always frightfully accurate. It will for sure destroy mankind.
|
"your experiences in using this groundbreaking tool?"...It blows...
|
AI is quite good for doing this ...
Discussion Summary
The discussion on the Audiogon Forum focuses on the utility and limitations of AI tools, particularly ChatGPT, in the context of High Fidelity (HIFI) audio systems. Participants share their experiences, highlighting both the benefits and drawbacks of using AI for audio-related advice, including system evaluation, gear comparison, and troubleshooting.
Key Points:
- Usefulness of AI: Many users found AI helpful for initial information gathering, equipment comparisons, and generating basic recommendations, appreciating its ability to provide quick and detailed information.
- Accuracy Concerns: A recurring concern was the potential for AI to generate inaccurate or misleading information, including incorrect specifications or illogical recommendations. Users emphasized the need to verify AI-generated content with other reliable sources.
- Subjectivity vs. Objectivity: AI tools were noted to struggle with the subjective nature of audio evaluation and personal preferences, often relying heavily on data without fully accounting for individual listening experiences.
- Learning and Improvement: While some observed that AI can learn and correct errors over time through user feedback, there was also a caution that AI might inadvertently incorporate user biases as facts.
- Other AI Tools: Besides ChatGPT, other AI tools such as GROK, Google Gemini, and Perplexity were also mentioned in the discussion.
Areas of General Consensus:
The overarching consensus is that AI tools can serve as a valuable starting point for audio-related research, offering quick access to information and comparisons. However, it's crucial for users to exercise critical judgment, verify the accuracy and relevance of AI-generated content, and not solely rely on it. AI is viewed as an assistive tool rather than a replacement for human expertise and personal listening preferences.
User Participation
- Total Users Contributed: 33 distinct users participated in the discussion.
- Users Who Made More Than One Post:
- emergingsoul
- dunham_john
- triton20trx
- signaforce
- duckworp
- guscreek
- User with Highest Participation:
- emergingsoul had the highest participation with 4 posts.
Here's what I will contribute
I suggest if people are going to take AI seriously they need to be as serious about understanding how it works and its limitations. Key points from my experience are:
Ask it how it generated an answer, especially if it sounds questionable, and you will learn how it works (assumptions, inference etc).
Chat GTP only commits some things to the memory of your account, and you can ask it to save things to memory, but there is a limit in the free version memory capacity. The memory can be edited. There memory applies to all chats.
Otherwise, Chat GTP only 'remembers' your content in individual chats and says each chat is independent. It is possible to put quite alot into one chat as broad and deep as you want to go. You can correct errors within the chat and it will be corrected on that basis because every response in a chat reviews all information in that chat.
It uses a combination of info its been trained with and search results and can infer opinion and marketing as fact, it can make many things sound quite plausible even if they are quite wrong. Its probably possible to get it filter info from search results such as "respond only using your training and do do incorporate search results". "Outline all relevant search results, its source and credibility." Then you can choose what info it should use in its response, such as "review your response to consider search results xyz"
Do not trust it, interrogate it, test results in multiple models and AI tools, use it to gain new information and perspectives to consider.
Learn how to prompt and be prepared to interrogate. Sometimes I start a new chat with a prompt based on lessons learned from a previous chat on the same subject that have improved the prompts and question I ask, and information I provide in the 2nd attempt for a better result.
I suspect the reason many are free is because they're learning from people's questions, probably to inform what training to have or capabilities to develop, or to better understand people's questions. I imagine there is a huge amount of computing capacity that is used by free accounts and there'd have to benefit to whoever is paying that cost I doubt its learning from or being trained by people's responses. Ultimately I suppose the objective is to motivate more people pay subscription .. people decide to graduate from limitations of free versions.
|
@dbastin
Well that was impressive. Curious to know what you doing in real life.
Would be interesting to better understand what the paid version offers, and I guess it shares memory between discussions which can be very valuable.
Would be very interesting to understand your system.
Thanks for offering a great summary of all that was contributed by so many herein.
|
If by 'real life' for a living, it's nothing to do with AI, programning or computing . I've learned about AI tools from researching, using and querying them. Beware if you ask AI to explain itself, because to specifically ask about the latest model.
Paid Chat GTP apparently has more personal memory and higher credit of tokens. Tokens are a measure computing resources. Some chats require more tokens, due to complexity etc. Neither memory or tokens are quantlified in something common like GB.
Gemini has no persistent memory but the free version seems to allow more tokens than chat GTP. Chat GTP seems a but more mature but Gemini is catching up very quickly. Co Pilot is a mixed bag, 3 different models/apps/interfaces that are not integrated together.
Do you mean my sound system or AI system?
The summary was Gemini responding to my prompt. My contributions were not generated by AI.
|
@dbastin
Yes I am interested in your audio system, and how it may have been inspired by AI
|
