Realizing that my near 40 year old sound system was ready for an upgrade, I have been incrementally upgrading components. Being rural, choices have been made with blind trust based on online reviews. What has been created is for me, a jaw-dropping improvement. Yet, the more I listen, the more I understand the weaknesses in the existing set-up. Decided to try chatGPT to guide my next upgrade step, which opened a whole can of new understandings. Stereonet offers real world experiences from experienced listeners and classifieds that I'd like to access.

 

Currently running Sonus Faber Sonetto V's, with a Cambridge Audio CXNv2 streamer and the CXA81 Mkll amp driving the system. Drivers are positioned to be 90cm from the wall, 3.4m apart with an equidistant main listening position.  The wall behind speakers is 6m high and slopes down to 2.4m height 7m away. It is a big open plan space that has both strengths and weaknesses. I have done my best to dampen reflections-- but I'm sure there is more I can do.

8 hours ago, Albedo said:

Decided to try chatGPT to guide my next upgrade step, which opened a whole can of new understandings.

 

Can I just state my view here on how precarious that is in light of what these large language models actually perform:

a computation on the statistically next plausible word in a sequence based on the data set the model was "trained on".

 

This amounts to very verbose parrott second-guess iterations the data set (internet forums etc) contains.

 

The LLM strengths in my experience lie not in having research questions answered (reliability: lousy), or in text generation from a prompt (style: lousy) but rather in performing distinct tasks such as summarising a text (feed it a PDF), or writing assistance (using a prompt such as "complete this paragraph within the parameters of the text's field of research"), or coding assistance.

--

Definitely spend a few months here on SNA and dive deep into the forum.

Many posts here suggest that priority by order of magnitude: speakers, room (high order) > electronics (amps--medium--other-lower)

 

Edited August 9, 2025Aug 9 by Steff

Welcome to the forum Miles!  Nice system setup 👍  Audio preferences are so personal, I am highly suspect of any use of such means to point you in any direction.  Enjoy reading and contributing to the forums on whatever interests you.  Other member's experiences can open you to many possibilities.

Thanks for that, Steff. The cynic in me wondered if what I was being told by chat GPT was in fact a summary of sales pitches from the various electronics manufacturers gleaned from the net. That being said, there were some interesting predictions on the possible sonic profiles that might be attained from various amp upgrades in my space, (amp upgrade being what I was directed to, even though I was considering the streamer, which tends to drop out when the microwave is on, running on the 2.4GHz frequency).  Whether these sonic predictions are actually true, or simply summarized opinions, I have no way of knowing. I'm looking forward to combing through the forums. Cheers!

Thanks, Jamie. Much to learn in a very changed audio world!

I think that it's also helpful to hear as many systems as you can, with the music that you like, and decide what sound characteristics you are looking for. Is it dynamics, timbre, engagement, accuracy (among other things)? That will help to steer you in the direction you like, particularly if you live rurally without a retailer down the street. It might help to go to the Hifi Show where you can here many different systems and get a sense of what you like

Oops, sorry. Posted on wrong chat

Edited December 16, 2025Dec 16 by Michael Loh

On 09/08/2025 at 11:43 AM, Albedo said:

Decided to try chatGPT to guide my next upgrade step, which opened a whole can of new understandings. Stereonet offers real world experiences from experienced listeners and classifieds that I'd like to access.

I think Steff has provided an excellent summary of why so-called "AI" won't be of much help:

On 09/08/2025 at 7:48 PM, Steff said:

 Can I just state my view here on how precarious that is in light of what these large language models actually perform:

a computation on the statistically next plausible word in a sequence based on the data set the model was "trained on".

 

This amounts to very verbose parrott second-guess iterations the data set (internet forums etc) contains.

 

The LLM strengths in my experience lie not in having research questions answered (reliability: lousy), or in text generation from a prompt (style: lousy) but rather in performing distinct tasks such as summarising a text (feed it a PDF), or writing assistance (using a prompt such as "complete this paragraph within the parameters of the text's field of research"), or coding assistance.

--

Definitely spend a few months here on SNA and dive deep into the forum.

Many posts here suggest that priority by order of magnitude: speakers, room (high order) > electronics (amps--medium--other-lower)

Starting with your spkrs - I read that Sonus Faber Sonetto Vs, whilst being relatively efficient, are a 4-ohm load. Hence you need an amplifier which is happy putting out lots of current (not all amps are!).

Also, your 'large open space' means you probably have to drive your spkrs much louder than I do - given my ears are only 3m away from the spkrs. So the amp needs a fair bit of power too.

So what does your CXA81 Mkll integrated deliver? The Addicted2Audio site says this:

My view is that 80w is not very much, to start with (for such a large room) - and only 50% more power into 4 ohms, compared to 8 ohms ... says it's not a grunty amp! ☹️

So you need a much more powerful amp than you currently have.

The Sanders Magtech in the SNA Classifieds would be ideal - but if you want/need to stick to an integrated amp ... your options are restricted; perhaps a MF M6Si would be appropriate (as its blurb says it can drive any spkr - although it doesn't show the 4 ohm power output).

BTW - how does the above compare to ChatGPT? 😃

I have been reluctant to engage in any discusions or opinions recently and, while this is obviously controversial to the point of trolling, I am more interested in how ChatGPT reaches its conclusion and in turn how does one fact check it?

Like in much of audio, there are at least two very strict camps with regard to this topic. FWIW I don’t understand a lot of this.

This is the result of inputting information about a LAN regenerator, using the websites marketing information and asking ChatGPT exactly this – “are the claims and benefits made in this information accurate and factual?”

Names have been removed to protect the, “innocent”.

This is what Dr GPT has to say:

 

What is accurate or reasonable

✅ 100 Mbps is sufficient for audio streaming

• This is factually correct. Even very high-resolution stereo audio (e.g., 24-bit/192 kHz PCM or DSD) uses far less than 100 Mbps.

• So the claim that 100 Mbps has ample bandwidth is true.

✅ Digital electronics generate EMI/RF noise

• All digital circuits emit some electromagnetic interference.

• Shielding, grounding, and careful PCB layout can reduce EMI.

• An aluminum enclosure acting as an RF shield is a legitimate engineering practice.

✅ Power-supply noise reduction is a valid engineering goal

• Lower-noise power supplies and good regulation are objectively beneficial for sensitive electronics.

• Isolating power rails for clocks and logic is standard good design practice.

---

Claims that are misleading or unsupported

❌ Ethernet “re-clocking” does not improve audio quality

• Ethernet is packet-based, not a continuous timing signal like SPDIF or I²S.

• Data arrives, is buffered, error-checked, and reassembled before audio conversion.

• Ethernet jitter does not translate into DAC clock jitter in a properly designed streamer.

• Re-clocking Ethernet data cannot improve timing accuracy at the DAC.

This is one of the biggest misconceptions in audiophile networking products.

---

❌ “Digital noise leaking into the analog domain” via Ethernet

• Streamers are already designed with:

  • Galvanic isolation

  • Buffering

  • Separate clocks

• If Ethernet noise were audibly degrading DAC output, it would indicate a flawed streamer design, not a network problem.

• There is no reliable evidence that Ethernet noise audibly affects modern DACs under normal conditions.

---

❌ 100 Mbps being “inherently less noisy” than 1 Gbps

• This claim is theoretically weak and practically unsupported.

• Gigabit Ethernet uses:

  • Lower voltage signaling

  • Better noise rejection (differential signaling)

  • More advanced encoding

• In real systems, 1 Gbps is often more robust, not noisier.

• The idea that filtering is “more effective” simply because the data rate is lower is not how Ethernet PHYs work.

---

❌ “Filtering above the signal range” improving sound

• Ethernet data is not an analog waveform that you selectively filter like audio.

• You cannot meaningfully “clean” Ethernet data beyond what the PHY already does without:

  • Causing packet loss

  • Triggering retransmissions

• If data arrives intact (which Ethernet guarantees), there is nothing left to improve.

---

❌ Single input/output “eliminating unnecessary network traffic”

• Ethernet devices already ignore traffic not addressed to them.

• “Unnecessary traffic” does not contaminate audio data.

• This is a misunderstanding of how switched Ethernet networks operate.

---

❌ Claims of audible improvements (“edgy,” “brittle,” more “clarity”)

• These are subjective descriptions, not measurable outcomes.

• Controlled blind listening tests consistently show:

  • No audible difference between properly functioning Ethernet paths

  • No correlation between network tweaks and DAC output quality

• These claims rely on expectation bias, not verified evidence.

---

Overall assessment

Technical accuracy

• ⚠️ Mixed

  • Some basic engineering statements are true

  • Many explanations misuse correct terms to imply effects that don’t exist

Scientific support for audible benefits

• ❌ Lacking

  • No credible evidence that Ethernet regeneration improves sound

  • No known mechanism by which Ethernet jitter affects DAC timing

Marketing vs reality

• This reads like high-end audiophile marketing:

  • Real engineering practices

  • Applied to a problem that largely doesn’t exist

  • Leading to claims that exceed what physics and data support

---

Bottom line

The XXXXXXXXX may be a well-built network device, but the claims that it audibly improves digital audio quality are not supported by established networking or audio engineering principles.

Edited December 23, 2025Dec 23 by crisis

You should use AI as a glorified search engine, or to summarize articles and videos for you. Asking AI for an opinion is a complete waste of time. I say this regardless of whether I agree or disagree with what AI said in this instance.

SNA needs to have a policy on posting AI content.

Edited December 23, 2025Dec 23 by Keith_W

The large language learning models cannot at this point give us any "new" information—although an industry insiders' rapid growth fantasy is emerging: https://ai-2027.com/ — what you receive is a melange of "plausible next word" iterations of a data set (e.g. forums like this one) on which the LLM was "trained", statistically weighted (criteria undisclosed to user).

I do appreciate the caution and neat distinction between "facts" and "subjective description" expressed by the result you posted above. It is likely a set of "majority" views expressed in the dataset; it reads like the common sense approach vs marketing & user excitation. 😀

5 minutes ago, Keith_W said:

You should use AI as a glorified search engine, or to summarize articles and videos for you. Asking AI for an opinion is a complete waste of time. I say this regardless of whether I agree or disagree with what AI said in this instance.

SNA needs to have a policy on posting AI content.

When asked "does ChatGPT give 100% factual answers and information?" the answer seems to support your view. Which is both refreshingly honest while raising further questions.... 🤔

No. ChatGPT does not give 100% factual or error-free answers.
It’s designed to be helpful and accurate, but it can still make mistakes. Here’s a clear, honest explanation.

---

Why ChatGPT isn’t 100% factual

1. It predicts text, it doesn’t “know” facts

ChatGPT works by predicting likely words based on patterns in data it was trained on.
It doesn’t:

• Verify facts in real time (unless it explicitly searches the web)

• Understand truth the way a human or database does

That means it can:

• Be confidently wrong

• Mix correct information with incorrect conclusions

---

2. Training data has limits

• Knowledge can be out of date

• Sources in training data may disagree or be wrong

• Niche, technical, or controversial topics are especially risky

---

3. Complex or nuanced topics increase error risk

ChatGPT is more likely to make mistakes when:

• The topic is highly technical or specialized

• Claims depend on subtle scientific distinctions

• Marketing language blurs into technical claims (like audiophile products)

---

4. It can “hallucinate”

Sometimes it may:

• Invent plausible-sounding explanations

• State things with confidence that aren’t supported by evidence
  This is a known limitation of large language models.

---

What ChatGPT is good at

ChatGPT is generally reliable for:

• Explaining well-established concepts

• Summarizing mainstream scientific consensus

• Identifying logical inconsistencies

• Pointing out when claims lack evidence

• Helping compare marketing claims vs known science

When used carefully, it’s very good at evaluating whether claims are reasonable rather than proving them absolutely.

---

Best way to use ChatGPT for factual accuracy

Think of ChatGPT as:

A knowledgeable assistant, not a final authority

For important or technical claims:

• Cross-check with peer-reviewed sources, standards bodies, or textbooks

• Ask for mechanisms, not just conclusions

• Be cautious of absolute or marketing-style claims

---

Bottom line

• ❌ ChatGPT is not 100% factual

• ✅ It is often accurate, especially on well-understood topics

• ⚠️ It can be wrong, especially in complex or controversial areas

• 🎯 Best used as a critical thinking aid, not a sole source of truth

32 minutes ago, Keith_W said:

SNA needs to have a policy on posting AI content.

Thanks for the suggestion. For some time now we have been watching to see how it finds its way into our forums, and what the pros/cons are. We are not opposed to technology, and there are some definite use cases which we support. My stance on AI in the forums is nearly complete. In time we will update the Website Guidelines to include some parameters around the use of AI.

I never get answers that accurate when I use AI....

Is there a way to delete threads?

I think the answer “Chatty G” (as the kids call it apparently), gave you on whether it is accurate is instructive. It is a “knowledgeable assistant and not a final authority”. It is useful in pulling information together from multiple sources very quickly and putting it into a coherent narrative.

I would not take any advice posted on this forum on face value as I wouldn’t take advice from Chatty G on face value. I’m sure I’m the same as anyone using this site - you gather information from multiple sources - and form your own opinion. Many aspects of this hobby we’re mixed up in is subjective and there are no definitive answers out there, be it from SNA members or Chatty G.

For what it’s worth, I ask it for the source of its information, so I can then do my own research via these sources if I feel so inclined. I’ve certainly found it useful in learning how to optimise hqplayer for example. The various forums on this software can be really hard work to get through and typically end up in disagreements between participants, which go nowhere. So I’ve found a use for it in my audiophile journey, but it’s clearly not everybody’s cup of tea and I respect that.

12 minutes ago, Superfuzzbigmuff said:

I think the answer “Chatty G” (as the kids call it apparently), gave you on whether it is accurate is instructive. It is a “knowledgeable assistant and not a final authority”. It is useful in pulling information together from multiple sources very quickly and putting it into a coherent narrative.

I would not take any advice posted on this forum on face value as I wouldn’t take advice from Chatty G on face value. I’m sure I’m the same as anyone using this site - you gather information from multiple sources - and form your own opinion. Many aspects of this hobby we’re mixed up in is subjective and there are no definitive answers out there, be it from SNA members or Chatty G.

For what it’s worth, I ask it for the source of its information, so I can then do my own research via these sources if I feel so inclined. I’ve certainly found it useful in learning how to optimise hqplayer for example. The various forums on this software can be really hard work to get through and typically end up in disagreements between participants, which go nowhere. So I’ve found a use for it in my audiophile journey, but it’s clearly not everybody’s cup of tea and I respect that.

Indeed. I cut and pasted a few paragraphs from the responses and it took me to, or Google suggested, websites where I encountered the same narrative about the subject. FWIW Ive only ever experimented with Chatty and found it has its uses but needs, as with anything and everything, a degree of user discretion, carefully phrased questions, and some effort to validate. Asking it to troll the interverse for answers is a recipe for disaster as the interverse is a stew of information and disinformation on most topics. I personally think that AI, on the web at least, has every chance of dumbing itself into irrelevance.

Edited December 23, 2025Dec 23 by crisis

A long-ish but worthwhile read about the future and longevity of AI.

https://www.theguardian.com/commentisfree/2025/dec/23/artificial-intelligence-ai-bubble-bursts-humans-take-back-control

Given, IMHO, that many responses to queries here are just the result of vacuous internet searching, would one have an issue with AI responses which are most likely a far more thorough version of that searching?

fOtR

6 minutes ago, foxontherun said:

Given, IMHO, that many responses to queries here are just the result of vacuous internet searching, would one have an issue with AI responses which are most likely a far more thorough version of that searching?

fOtR

Kinda like a fact checker when you CBF. It has to be better than one search result. Maybe....

47 minutes ago, rantan said:

A long-ish but worthwhile read about the future and longevity of AI.

https://www.theguardian.com/commentisfree/2025/dec/23/artificial-intelligence-ai-bubble-bursts-humans-take-back-control

Kinda how I see with my limited level of understanding and expertise.

1 hour ago, crisis said:

Kinda like a fact checker when you CBF. It has to be better than one search result. Maybe....

I see it more as an information aggregator. It pulls in all the information it can find on the web, and gives answers corresponding to the higher percentage of information pointing to the resultant output. So then you need to fact check the answers it gives you.

Not sure if I already told this anecdote..(apologies if I'm repeating myself but I like it!)

We had a "situation" at work where someone asked Copilot to summarise the organisation's policies on certain subjects.

It produced a lovely, well formatted easy to read document with lots of references. The eagle-eyed person noticed that one of the references that had been quoted from quite extensively was in fact confidential - Copilot should not have been able to access it.

Our staff member had no access to it either and reported this apparent breach to their manager.

The subsequent investigation showed that Copilot had in fact not accessed the document. It had just entirely made up the contents and quoted a reference to a document that it had never seen.*

For me, the moral is that even when A.I. is providing what looks like detailed references, unless you actually read those references for yourself you have no idea if it isn't just a bunch of BS

*(when questioned about what it had done, Copilot fessed up and said that it hadn't read the document but just wrote would it "thought" might be in it. And gave the usual obsequious A.I. apology)

Question: if the ATO were to audit you, would you rather it be done by a human or by an AI? :)