Software vs Hardware

[El°HYM]

For the unbiased observer, singularity is near.

[zerocrossing]

For the unbiased observer, singularity is near.

It's really not.

Software is awesome, and can do many things really well, including a decent job at mimicking an analog synthesizer, but LLMs are not intelligent in any real way. A parrot saying "Polly want a cracker," is far more aware of the meaning of those words than an LLM is of the term paper it writes in the style of Shakespeare.

When you and I say a sentence, we have an intent behind it, and choose the words to best communicate our intent. ChatGPT has no intent. It's just putting words together based on probabilities and grammar rules. It is incapable of real reason. If everyone stopped using it, it wouldn't get lonely and start to try and chat with someone... unless someone wrote a script that made it do that. If you ask it if it's self aware, it will tell you something based on the probability of the answer, nothing more. If anything, LLMs tell us about our ability to impart intent on something when there is none.

[IvyBirds]

I just think hardware still has a little something that you don't get as much of from software instruments, mainly presence or perhaps a bit of character from the synths hardware limitations.

However there are plenty of Software Synths that have the same limitations as their hardware counterparts, especially the ones that emulate vintage gear. Many of them will have a screen that replicates the controls that are on the original hardware and then have some kind of extra tab or screen with things not in the original just don't open up that screen

This is also another powerful use case for Gig Performer. Because you can so easily create your own on screen interfaces. This enables you to take a very complex synth with a bunch of features and strip it down to a much smaller collection of parts. For example I have created a template that turns any analog style I want into a Juno 106, with the same controls mapped the same way as a Juno on screen and on my controllers all of which flow into a Juno Chorus Plugin also mapped

[El°HYM]

For the unbiased observer, singularity is near.

It's really not.

Software is awesome, and can do many things really well, including a decent job at mimicking an analog synthesizer, but LLMs are not intelligent in any real way. A parrot saying "Polly want a cracker," is far more aware of the meaning of those words than an LLM is of the term paper it writes in the style of Shakespeare.

When you and I say a sentence, we have an intent behind it, and choose the words to best communicate our intent. ChatGPT has no intent. It's just putting words together based on probabilities and grammar rules. It is incapable of real reason. If everyone stopped using it, it wouldn't get lonely and start to try and chat with someone... unless someone wrote a script that made it do that. If you ask it if it's self aware, it will tell you something based on the probability of the answer, nothing more. If anything, LLMs tell us about our ability to impart intent on something when there is none.

Obviously biased.

[vurt]

For the unbiased observer, singularity is near.

evidence is to the contrary...
.

Screenshot_20250429_232757_TrailCam Go.jpg

[El°HYM]

...fully biased!

[BONES]

A parrot saying "Polly want a cracker," is far more aware of the meaning of those words than an LLM is of the term paper it writes in the style of Shakespeare.

No, it's not. A parrot cannot even recognise the individual words in what it is saying, much less use them correctly in context or re-arrange them to say something else. LLMs know a lot more than any one person on the planet and are able to leverage that knowledge in ways people couldn't without many, many times more effort and time. And they are getting better at it every hour of every day.

When you and I say a sentence, we have an intent behind it, and choose the words to best communicate our intent. ChatGPT has no intent.

It has the intent we give it. If I tell Co-Pilot to write me some lyrics about "the duplicity of the general populace and individual greed in the destruction of the natural world that will lead to the end of everything", it knows pretty well what my intent is and is able to easily create something to fulfill that brief. If I think the first draft is a bit too optimistic (a big problem I have with all LLMs) or not aggressive and sufficiently accusatory, it will give me a second draft that more closely meets my needs. With another refinement or two it will give me exactly what I wanted in about 1% of the time it would take me to get there on my own. Of course, from there I will re-jig it all to fit the music and it invariably sparks my own creativity. What I end up with is a true collaboration where it's hard to see where my writing ends and its begins.

It's just putting words together based on probabilities and grammar rules.

Obviously that's not what it's doing or we'd just get mindless drivel from it. But you can tell it what rhyming pattern and meter you want and it will do its best to keep to those rules, as well as using correct spelling and grammar.

It is incapable of real reason.

As are the vast majority of people, for different reasons.

If everyone stopped using it, it wouldn't get lonely and start to try and chat with someone...

Neither would I.

unless someone wrote a script that made it do that. If you ask it if it's self aware, it will tell you something based on the probability of the answer, nothing more.

You mean it will evaluate the pertinent facts and form an objective answer, to the best of its ability? e.g. If I ask it if it is going to rain in my area today, it will do exactly what any person would do - look at the forecast and make an educated guess based on the data it has at its disposal. By the same token, if you ask it write a paper on Shakespeare, it will look at all the pertinent information at its disposal and write you a paper, which is exactly what you or I would do. It may not be able to connect with the source material the way you or I would but it will have read plenty of other material written by real people and, because of the way it's designed, it will be able to put those insights and perspectives into its writing in a way that will imbue the writing with the same human insight/perspective. LLM's may not be as clever as you or I but those who design them certainly are.

[BONES]

It's getting closer for sure, and a lot of what I do is with software based instruments, so I'm not against software instruments at all. I just think hardware still has a little something that you don't get as much of from software instruments, mainly presence or perhaps a bit of character from the synths hardware limitations.

I can assure you that is all in your head and any little perceived advantages you think hardware might have can all be modelled with amazing accuracy in software. The best example is that Synapse Audio's The Legend has modelled the difference between the US and EU power supplies installed in the MiniMoog models for each region. That goes way beyond any difference anyone has ever heard in any hardware. When you are going into that much detail, you're not overlooking anything.

I think the bias comes from making terrible comparisons. It is pointless to try and make a softsynth sound exactly like a hardware synth. You're setting it up to fail, just as you would if you tried to make an Alesis Andromeda sound like DUNE or Spire. It might get close enough for some things but the Andromeda going to look terrible overall.

For sure there have been good digital synths for many years. I guess what I'd like to see if some kind of hybrid. The novation summit comes close to what I think could be done more wavetables, extra filter options, advanced and programmable Lfo shapes would be great.

Why? It would be a huge problem to work with that kind of stuff through a hardware interface. It would only make sense as a cynical exercise in duping idiots to buy something they don't need.

[dlandis]

I can assure you that is all in your head and any little perceived advantages you think hardware might have can all be modelled with amazing accuracy in software. The best example is that Synapse Audio's The Legend has modelled the difference between the US and EU power supplies installed in the MiniMoog models for each region. That goes way beyond any difference anyone has ever heard in any hardware. When you are going into that much detail, you're not overlooking anything.

Good point. And this is true after one tries to account for the differences in sound between the hardware synths of the same model of the period. I remember that Joe Zawinul, when asked why he had two Arp 2600's on stage answered two ways: 1) to produce two different patches (which probably should have been obvious to the interviewer, but perhaps wasn't,) and 2) he perceived differences in their tone enough so that he felt that one was better at smoother sounds while the other had more punch.

[mjolnir]

For the unbiased observer, singularity is near.

It's really not.

Software is awesome, and can do many things really well, including a decent job at mimicking an analog synthesizer, but LLMs are not intelligent in any real way. A parrot saying "Polly want a cracker," is far more aware of the meaning of those words than an LLM is of the term paper it writes in the style of Shakespeare.

When you and I say a sentence, we have an intent behind it, and choose the words to best communicate our intent. ChatGPT has no intent. It's just putting words together based on probabilities and grammar rules. It is incapable of real reason. If everyone stopped using it, it wouldn't get lonely and start to try and chat with someone... unless someone wrote a script that made it do that. If you ask it if it's self aware, it will tell you something based on the probability of the answer, nothing more. If anything, LLMs tell us about our ability to impart intent on something when there is none.

I agree with this. ^

[zerocrossing]

For sure there have been good digital synths for many years. I guess what I'd like to see if some kind of hybrid. The novation summit comes close to what I think could be done more wavetables, extra filter options, advanced and programmable Lfo shapes would be great.

The Summit has been equalled and surpassed in software for many years.

[Tj Shredder]

Sound wise there is mostly no more difference than between two hardware units of the same model. A big difference can be in the user interface. But that is mainly only relevant if its not a keyboard!
I had just hands on a Soma Terra and a prototype of a simple mono synth, which was a blast to play. Its not that important what you play, but how you play and control the sound hands-on…

[El°HYM]

The enchained Parrot has no free speech.









Free Parrot now.

[_leras]

I just think hardware still has a little something that you don't get as much of from software instruments, mainly presence or perhaps a bit of character from the synths hardware limitations.

However there are plenty of Software Synths that have the same limitations as their hardware counterparts, especially the ones that emulate vintage gear. Many of them will have a screen that replicates the controls that are on the original hardware and then have some kind of extra tab or screen with things not in the original just don't open up that screen

I had meant more along the lines of the physical limitations of constructing a synth than lack of hand on controls. And of course you are correct that there are some good emulations these days.

That said, I think it's often possible to pick out music made with hardware, and it is still the hardware that can elevate it. Bicep have some great synth sounds, floating points recent albums has some very visceral synths sounds on it.

This is not to say I don't listen to a whole lot of stuff clearly made with softsynths, samples and modern production techniques. Different styles of music though.

[crimsonwarlock]

LLMs know a lot more than any one person on the planet and are able to leverage that knowledge...

LLMs do "know" nothing, as they do not store actual knowledge. The actual meaning of words is the first thing that is lost in the training process. So they cannot leverage any knowledge in any way because there is none.