OpenAI ChatGPT has some interesting insights about audio

[Koo]

I've been playing with the new ChatGPT bot by OpenAI and it's crazy how well it can explain various concepts. Here is the link if you want to try it out. It's currently free and open for everyone during the testing phase https://chat.openai.com

Here are some examples:

This could be a really good tool to compress information from various sources and help learn new things easily. I would like to see them add support for actual sources where the information is from.

Here is an example where the AI went off the rails.. or maybe it knows something we don't?

 

[kemmler3D]

I think it also suffers from the same thing as Meta's glacticus (or whatever they called it), where it always says something pretty plausible and well-phrased, but also often says something wrong, or almost right.

Here's an example:

This is a bit misleading at best.

Or this, which doesn't fully grasp the concept and just kinda spews relevant word salad... I'm confused anyway:

I think the huge problem with these AI text generators... they can easily SOUND correct and it spits out text that seems intelligently written. But they sound correct while saying stuff that is wrong, almost right, or just doesn't make sense. And so it's easy for someone to read this, assume it's right because it sounds right, and end up learning something completely wrong.

 

[dadregga]

I think it also suffers from the same thing as Meta's glacticus (or whatever they called it), where it always says something pretty plausible and well-phrased, but also often says something wrong, or almost right.

Here's an example:

View attachment 247833
This is a bit misleading at best.

Or this, which doesn't fully grasp the concept and just kinda spews relevant word salad... I'm confused anyway:

View attachment 247834

I think the huge problem with these AI text generators... they can easily SOUND correct and it spits out text that seems intelligently written. But they sound correct while saying stuff that is wrong, almost right, or just doesn't make sense. And so it's easy for someone to read this, assume it's right because it sounds right, and end up learning something completely wrong.

Should be perfect for turning out audiophile marketing copy then.

 

[Zedly]

Unfortunately, some of the audiophile propaganda was in its training database.

 

[antcollinet]

I've been playing with the new ChatGPT bot by OpenAI and it's crazy how well it can explain various concepts. Here is the link if you want to try it out. It's currently free and open for everyone during the testing phase https://chat.openai.com

Here are some examples:
View attachment 247810

View attachment 247811
View attachment 247812

This could be a really good tool to compress information from various sources and help learn new things easily. I would like to see them add support for actual sources where the information is from.

Here is an example where the AI went off the rails.. or maybe it knows something we don't?

View attachment 247813

In your last question/answer, the problem is not in the answer, but in the undefined nature of "audio quality"

 

[kemmler3D]

In your last question/answer, the problem is not in the answer, but in the undefined nature of "audio quality"

That's pretty debatable, since the most appropriate answer would be "they don't, usually". The "you can only hear it if you have special ears" tag at the end was also especially pernicious in this case.

Overall I think these models don't seem to have enough reasoning ability to tell fact from hogwash, although they clearly have the ability to SEEM like they have that ability, which is really dangerous. It's like being able to build a bridge that looks completely sturdy and is actually a deathtrap.

 

[antcollinet]

OK - lack of a common agreed definition between the asker, and the answerer.

And yes, I agree - they are only as good as the text they are trained on. I wonder where that came from. If a "representative" selection from the world of audiophilia, not that surprising that it mirrors the wrongness that exists therein.

 

[soundwave76]

Yep, I am liking ChatGPT more and more...

 

[soundwave76]

The story continues...

 

[pma]

It seems it is able to find a reply on a common subject of human consumption or to write a marketing essay on such subject, however a question to a unique technology gives repeatedly an answer like this:

 

[soundwave76]

Riffusion

'This is the v1.5 stable diffusion model with no modifications, just fine-tuned on images of spectrograms paired with text. Audio processing happens downstream of the model.'

 

[kemmler3D]

The audio generation is better than I expected but still sounds as much like FFT artifacts as music.

What's interesting about their "morph" between certain sounds is that it echoes much earlier work on additive synthesis, (I want to say it was in the 80s or something) where individual harmonics were gradually shifted between the timbres of two instruments. While this seems like it ought to be a pretty unique-sounding, mind-bending kind of effect, the end result is more like a simple cross-fade. I think this is a small step beyond that, but still doesn't sound like anything is transforming into anything else, to me.

I wonder how far this technique can go? I wonder if they can train the model to sound truly realistic this way, since spectrograms have a trade-off between time and frequency resolution.

 

[pseudoid]

Would I be way-OT to recommend the following write-up about "Without Consciousness, AIs Will Be Sociopaths"(2023/01/14)... just as food-for-thought?
imo: Humans are in need of TuringTest v.2035.

 

[Willem]

I am glad I am no longer responsible for the integrity of university exams. First we had plagiarism, next we had the ghostwriter problem (I participated in a national tv program where I successfully identified all of those - it was easy) but this looks like a harder challenge. It still is exam fraud.

 

[pseudoid]

It still is exam fraud.

Even animals "cheat'.
During a job-interview, this young-ish EE (w/a good CV) shared with me how much of a team-player he was by describing to me that he "assisted" other fellow students with their coursework.
Shocked; I asked what kinds of coursework assistance? Needless to say it was a "no hire" when I was told "...with their assignments..."

I am not certain if he was related to George Santos....

 

[stunta]

Write a sleazy sales pitch for an audiophile snake-oil device



Are you tired of mediocre sound quality from your audio devices? Do you want to take your listening experience to the next level? Introducing the ultimate audiophile solution: our snake-oil device.
This little device may look small, but it packs a powerful punch. It is designed to enhance the audio signal and eliminate distortion, resulting in a crystal clear and immersive sound. Not only will your music sound better, but you will also hear details in your favorite songs that you never knew existed.
But that's not all, our device is also made with the highest quality materials and is built to last. It's compatible with all audio devices, from your phone to your high-end audio system.
Don't settle for less, upgrade to the ultimate audiophile solution and elevate your listening experience today. Limited stock available, so act fast!
Note: Snake-oil is a term used to describe products that are marketed with false or misleading claims and have no real benefit. I do not encourage the use of such products and this text is just an example of a sleazy sales pitch.

 

[amirm]

Just set up an account with them to see if it can clean up my writing instead of making up answers. I was pleasantly surprised that it can do that. That said, it did change the meaning a bit and took out some specifics. Let me know whether you think my version is better or not. It is the conclusion in my iPurifier3 review:

My writing:
It seems that the iPurifier does what it says it does: clean up the signal/power lines over USB and with it help devices that need it. Then again every DAC I recommend of which there are probably 50 to 100, already perform their own filtering and don't need any help. After all, I test them all on my desktop workstation without any such filters. Adding an active device to the USB connection can reduce available power and potentially decrease measured performance as show in the case of Schiit Modi 3+. Seeing how a better DAC costs less than iPurifier3 anyway, there is no reason to deploy such a device.

ChatGPT:
The iPurifier is a device that claims to clean up the signal and power lines over USB, which can improve the performance of certain devices that may be sensitive to electrical noise. However, many high-quality digital-to-analog converters (DACs) already have built-in filtering and do not require additional help. Additionally, adding an active device like the iPurifier to the USB connection can decrease available power and potentially decrease overall performance. Given that a better DAC can often be obtained for less cost than the iPurifier, there may be little reason to use such a device.

 

[DonR]

It seems to provide a more formal presentation rather than your original. I actually used it to write a legal document and it worked quite well.

 

[amirm]

I noticed that too.

 

[TonyJZX]

ironically Microsoft is the pioneer on this? They had Tay more than 5yrs ago.

Tay (chatbot) - Wikipedia

en.wikipedia.org

This ChatGPT also passed the bar exam.

https://edition.cnn.com/2023/01/26/tech/chatgpt-passes-exams/index.html

From Wharton too. Which is famous for being the college that passed esteemed President and document historian Donald Trump.