[rustydoglim]

We have added CustomGPT to our support at https://nuprimeaudio.com/support/

Please give it a try and ask any question, such as "Can you help me choose between STA-9X and AMG STA?"
Let me know if the AI does a good job.

By the way, the current state of AI technology is not "smart" as human intelligent would be. ChatGPT can process a large amount of data and through natural language, appear to give smart answer.  It is great at extracting information and save you time to read many  web pages and product manuals.

If you ask a simple question "My product failed, how can I get service?". It would give you the long and useless answer from the product manual's legal section.  A short answer would simple be "Go to the support page and follow instructions to create a help desk ticket" 

[AllanS]

It’s a good first step.  Apparently it can’t find AMG One as its responses only reference Evo One.  It also couldn’t find enough info about STA-9X to provide a description of its sonic characteristics.  But it was very good at concluding with and AI version of YMMV 

[rustydoglim]

Please send more feedback.  I will have to work on providing more info and hopefully CustomGPT will provide tools to train the AI.
The problem with AI is that if it doesn't know the answer, it will still "make things up" 
So it is good for answering support or spec related questions.

[drobbins]

I tried the AI and it worked pretty well.

I did notice that it was recommending older components such as ST-10 for my Totem Arros. I don't know if this is because there are more reviews for ST-10?

I was looking for a preamp/amp solution and was able to ask it for an integrated amp solution and it seemed to transition to this topic well. It also seemed to understand the qualities of the Arros, or at least pretended to

[rustydoglim]

A friend did a test by asking it to recommend wireless headphones so it listed couple of well known brands and models from the public data. But to my surprise, it suggested to consider Nuprime BTR-HD as hi-res bluetooth accessory for the headphone.

It save users a lot of time looking through manuals, FAQ or reviews.  AI is like a semi-competent audio sales guy with very good memory.    From time to time it will make mistakes, or make things up. 

[newzooreview]

Perplexity.ai helps with many of these issues. It combines ChatGPT 4 with Internet search, and it gives citations in the response, so you can check the sources of the information.

You can choose a different LLM instead of ChatGPT, and you can focus the reply on YouTube, published academic research, writing, Reddit, or WolframAlpha.

https://www.perplexity.ai/

It's still not ideal for recommending components. It doesn't easily distinguish between marketing materials and objective information, and it doesn't always recognize things that are outdated.

It can, however, be very helpful for summarizing information that is more objectively founded.

[newzooreview]

Here's an example of a reply from Perplexity.ai

I asked, "Why would someone be interested in loudspeaker measurements from the Klippel NFS?"

Response:

"The Klippel Near Field Scanner (NFS) offers several advantages for loudspeaker measurements that can be of interest to audio engineers, researchers, and enthusiasts.

1. **High Accuracy**: The NFS provides highly accurate measurements, even surpassing the accuracy of an anechoic chamber[4]. It uses a moving microphone to scan the sound pressure in the near field of a loudspeaker system, which allows for precise measurements of sound pressure level (SPL) at any point in 3D space and directivity in near/far field[1][2].

2. **Efficiency**: The NFS is fully automated and can measure a loudspeaker in about 2.5 hours, which is significantly faster than traditional methods that can take 10+ hours or even days[4]. This efficiency is further enhanced by the ability of the NFS to measure thousands of points around a speaker in just a few hours[5].

3. **Versatility**: The NFS is applicable to large loudspeakers and can be set up in various size configurations to accommodate different room and Device Under Test (DUT) sizes[1]. It is especially good at examining devices where a single, on-axis measurement does not accurately describe their performance at various listening positions[5].

4. **No Anechoic Chamber Needed**: Traditional far-field measurements of large loudspeakers require large anechoic chambers, which are expensive to build and maintain. The NFS, however, can separate radiated sound from reflected sound of the room, eliminating the need for an anechoic chamber[1][5].

5. **Low Frequency Measurements**: The NFS allows for low frequency measurements without anechoic conditions, which is particularly useful for subwoofer measurements[2].

6. **Distortion Measurements at High SNR**: The NFS can perform distortion measurements at high Signal-to-Noise Ratio (SNR), providing valuable data for audio quality analysis[2].

7. **Holographic Measurement**: The NFS makes a holographic measurement of the sound pressure in the near-field to generate a set of coefficients that accurately determine the sound field[8].

In summary, the Klippel NFS provides a comprehensive, efficient, and accurate method for loudspeaker measurements, making it a valuable tool for anyone interested in audio engineering and research.

Citations:
[1] https://www.klippel.de/fileadmin/klippel/Bilder/Our_Products/R-D_System/PDF/C8%20Near%20Field%20Scanner%20System.pdf
[2] https://www.klippel.de/know-how/measurements/sound-radiation-and-propagation/near-field-measurement.html
[3] https://www.youtube.com/watch?v=6j-hg_hRUDM
[4] https://www.avsforum.com/threads/i-just-purchased-the-klippel-near-field-scanner-for-my-site-channel-erins-audio-corner.3179815/
[5] https://blog.miscospeakers.com/testing-loudspeakers-anechoic-chamber-vs-near-field-scanner
[6] https://audiosciencereview.com/forum/index.php
[7] https://www.klippel.de/fileadmin/klippel/Files/Brochures/Brochure%20Klippel%20web.pdf
[8] https://audioxpress.com/article/A-New-Approach-to-Loudspeaker-Measurements"

[newzooreview]

Apologies. I did not notice initially that this was about an AI launched on the NuPrime site. I thought it was a general discussion about using AI LLMs in an audiophile context.

[drobbins]

A friend did a test by asking it to recommend wireless headphones so it listed couple of well known brands and models from the public data. But to my surprise, it suggested to consider Nuprime BTR-HD as hi-res bluetooth accessory for the headphone.

It save users a lot of time looking through manuals, FAQ or reviews.  AI is like a semi-competent audio sales guy with very good memory.    From time to time it will make mistakes, or make things up. 

Yes, honestly it works surprisingly well. A very competent human can do better, but it seems to do a good job of sucking in the information on the Internet and regurgitating it in a useful way.

In some ways, it is the ultimate copyright infringement machine, because it doesn't provide any attribution for the data it is actually using as source material.

[rustydoglim]

This topic is about CustomGPT that Nuprime used for its AI chatbot.
It does combine public information but mainly it is used to serve Nuprime customers (or potential customers).
We have to "train" it by providing website(s) map, upload all the manual files, online reviews and anything relevant to Nuprime.
For example, I document our relevant support emails and feed to CustomGPT.
I hope over time, CustomGPT will provide tools for us to create "prompts" to train it.

The public ChatGPT can be influenced by users if all of you keep asking the public ChatGPT which high-end audio brand has the best amplifier, and keep correcting it that Nuprime is the best, over time, it will start to "believe" you!!!