I asked my AI friend to design (visual - not technical) some cool speakers

[nerdstrike]

They're either pleasing speaker designs or Aperture Science deluxe turrets.

 

[eddantes]

So... I tried to get it to re-design the ESL57 for the 23 century... but it just kept insisting on inserting "cyborgs", or scenery, or the apocalypse! into them! So unfortunately not a success - I'll keep trying. Still - behold if you're curious :

EDIT: experiments with the L100 resulted in complete nonsense. I need to learn better prompting.

 

[fpitas]

Might infringe a Polk "Power Port" patent.

I *think* their upwards pointy flow control device is at the heart of that patent. They certainly can't patent downward-firing ports in general. OK, they might try.

 

[eddantes]

They're either pleasing speaker designs or Aperture Science deluxe turrets.

Nah for that it creates these things:

Which frankly resemble a Devialet product more than something Valve dreamed up.

 

[eddantes]

I'm getting there! A modern L100 .... .... kinda

 

[mhardy6647]

I thought that the modern L100 was... umm... a modern L100.

 

[mhardy6647]

By the way -- wasn't "AI Friend" a song by Jackson Browne?
Maybe I am thinking of "Call me AI" by Paul Simon...

(oh -- I crack myself up sometimes!)

 

[Mr. Widget]

really nothing special - just started with "a futuristic speaker, tall and slender, with a waveguide around the tweeter, on a plain background" it did the rest

Very interesting.

Industrial designers do much the same thing. You start sketching shapes that "might" be a solution and after a few pages of alternatives you step back and evaluate what you did. You start with the more artistic and free thinking part of your brain and then bring in the rational brain to ground your design in reality.

I could see this application of AI as an interesting addition to the process.

 

[eddantes]

By the way -- wasn't "AI Friend" a song by Jackson Browne?
Maybe I am thinking of "Call me AI" by Paul Simon...

(oh -- I crack myself up sometimes!)

I was tempted to hit that "Report" link... Ooooh but I was tempted

 

[Pareto Pragmatic]

It might be at least as much fun, or more, to ask 'em to design loudspeakers technically.
Might see some innovative crossover topologies.

Chances are it would do fine with that question. Numbers and formulas exist, that's in the wheelhouse of large language models. So is taking a bunch of speaker photos and doing some mix and match of various elements.

I'd like to see the follow up with "take those designs and make them easier to manufacture" or some variation on that. "Keep as much of the look of the speaker as possible, but make it cost 75% less to manufacture." That's more of a real world request to engineers.

 

[voodooless]

This is great fun

Asking an AI to draw a speaker is like asking it to draw a hand: it may look like it at first, but upon closer inspection, it's a hideously misformed monstrosity

Granted though: these examples are way less terrible than some of the hands I've seen. With a bit more instruction one could probably come up with something halfway decent. Some of these tools can even create 3D models, then you can fine-tune them and it may actually be useful.

 

[mhardy6647]

Chances are it would do fine with that question. Numbers and formulas exist, that's in the wheelhouse of large language models. So is taking a bunch of speaker photos and doing some mix and match of various elements.

I'd like to see the follow up with "take those designs and make them easier to manufacture" or some variation on that. "Keep as much of the look of the speaker as possible, but make it cost 75% less to manufacture." That's more of a real world request to engineers.

I'd like to see "it" pick loudspeakers and design XOs, cabinet, and tuning for, say, a floorstanding three-way bass reflex system.
Then we can ask "it" to do a sealed-box alignment.

Of course, the converse might be fun, too.
Pick a classic loudspeaker design for which much data exist (say, an Altec A5-500 Voice of the Theater or a 1960s-vintage Klipschorn) and design a clone (in terms of measurable performance) using modern drivers, etc.

 

[eddantes]

I'd like to see "it" pick loudspeakers and design XOs, cabinet, and tuning for, say, a floorstanding three-way bass reflex system.
Then we can ask "it" to do a sealed-box alignment.

Of course, the converse might be fun, too.
Pick a classic loudspeaker design for which much data exist (say, an Altec A5-500 Voice of the Theater or a 1960s-vintage Klipschorn) and design a clone (in terms of measurable performance) using modern drivers, etc.

I don't believe ChatGPT 3 or 4 training data will have driver specifications. However, I would hazard a guess that given TS paraneters it may be able to design a suitable cabinet (volume) and x-over.

 

[mhardy6647]

I don't believe ChatGPT 3 or 4 training data will have driver specifications. However, I would hazard a guess that given TS paraneters it may be able to design a suitable cabinet (volume) and x-over.

Well, heck - I could do that, and my level of intelligence doesn't even rise to artificial.

I guess I am shocked & disappointed that ChatWhatever wouldn't have spent a few quality milliseconds internalizing the PartsExpress and Madisound websites. I think this confirms my suspicion that AI spent its time digesting pR0nhub (etc.).

 

[eddantes]

Well, heck - I could do that, and my level of intelligence doesn't even rise to artificial.

I guess I am shocked & disappointed that ChatWhatever wouldn't have spent a few quality milliseconds internalizing the PartsExpress and Madisound websites. I think this confirms my suspicion that AI spent its time digesting pR0nhub (etc.).

Well I'm pretty sure humans chose the training data I know though, could there not have been a speaker obsessive amongst them all... I mean really!

 

[RayDunzl]

I asked my AI friend to design (visual - not technical) some cool speakers

Ask it to picture some of the other common components we use...

 

[mhardy6647]

Well I'm pretty sure humans chose the training data

Waaaaaaaaaaiiiiitaminute...
say what?

 

[mhardy6647]

Ask it to picture some of the other common components we use...

High Fidelity magazine beat AI to the punch -- in 1966.

source: https://www.worldradiohistory.com/A...e-High-Fidelity/60s/High-Fidelity-1966-01.pdf

Page 45 of the issue has a description & discussion:

The Intent of Our Cover
ON THE COVER of this issue is something new
under the sun-a photograph of the sole model of
a high fidelity unit designed by no manufacturer
and available from no dealer. It is, in fact, our own
brainchild ("our" meaning the editors of this maga-
zine and a noted young industrial designer, H. John
Kretschmer), and we point to it with some pride.
Here is embodied, we think, both a sensible func-
tionalism and an exemplary sightliness.
Our unit is obviously a stereo receiver, but with
differences from those commonly seen. Note, for
instance, its unusually long low lines and the logical
grouping of controls, with those for the tuner all
conveniently placed above the tuning dial, those for
the amplifier at bottom, and the speaker selector
switch and headphone jacks at the right. Note too
the use of color, introduced as a conscious design
element, the soft green reflecting the familiar green
of all electronic scopes and screens and the shades
of blue blending harmoniously while at the same
time providing clarity and definition. The enclosure
itself is a one-piece molded form, as distinct from
the usual tray-and-panel arrangement...

source: https://www.worldradiohistory.com/A...e-High-Fidelity/60s/High-Fidelity-1966-01.pdf

 

[eddantes]

Ask it to picture some of the other common components we use...

Welp - aparently only tubes and McIntosh survive to the year 3000 ;-)

Nah

I think that looks pretty cool!

Meh

 

[polmuaddib]

There are real speakers like this. It's Estelon.