I asked AI to design a bass cab

[41hzthunder]

I've seen AI scan Talkbass for answers on speaker questions, whether right or not. It doesn't have the ability to distinguish a good reply with a bad one. I believe it does look at number of views when weighing the value of data. Why are good answers not more popular?

 

[Al Kraft]

I've seen AI scan Talkbass for answers on speaker questions, whether right or not. It doesn't have the ability to distinguish a good reply with a bad one. I believe it does look at number of views when weighing the value of data. Why are good answers not more popular?

IMO,...most good answers in engineering are "it depends", and folks want instant answers not a couple minute discussion to help narrow down specifics of their use case so the situationally relevant logic/math/science can be applied. Good answers take time, people value fast and easy...that's my best guess.

 

[rockear]

It is kind of crazy that they seemingly haven't programmed these things to have any kind of discretion vis a vis weighting one data source against another. It's like they just take every opinion and shake them up in a barrel. You might think they'd be able to figure out "This horse guy seems to know a thing or two about a thing or two... I wonder what he does for a living?"

 

[fdeck]

It is kind of crazy that they seemingly haven't programmed these things to have any kind of discretion vis a vis weighting one data source against another. It's like they just take every opinion and shake them up in a barrel. You might think they'd be able to figure out "This horse guy seems to know a thing or two about a thing or two... I wonder what he does for a living?"

People are doing this, but it's not necessarily available to the general public or for free. There's a growing market for AI products that are customized or specialized. One reason is that when data are filtered by some kind of vetting process, the resulting data set becomes something of potential value that can be sold, or used to ones' own advantage. The AI companies sell "enterprise" service where they promise that they won't store or use their customers' proprietary information.

Remember the saying, "if a product is free, you're the product."

 

[Al Kraft]

People are doing this, but it's not necessarily available to the general public or for free. There's a growing market for AI products that are customized or specialized. One reason is that when data are filtered by some kind of vetting process, the resulting data set becomes something of potential value that can be sold, or used to ones' own advantage. The AI companies sell "enterprise" service where they promise that they won't store or use their customers' proprietary information.

Remember the saying, "if a product is free, you're the product."

Well said IMO (and at no extra charge).

 

[Luther]

My requirements were light, loud and small enough to fit in an airline overhead. I asked for a 6o to 3.5k hz frequency response.

Invalid Link Removed

The design looks pretty good to me.

Comments?

I asked AI to buy one for me.

 

[wimpy]

and is there any TBers who have already implemented the construction according to the AI plan? I feel the urge, but I'm trying to fight this handicap of mine

 

[fdeck]

and is there any TBers who have already implemented the construction according to the AI plan? I feel the urge, but I'm trying to fight this handicap of mine

I've plugged the numbers into my modeling program but didn't save the results. I wouldn't buy materials before doing that.

 

[No such thing]

From Claude Sonnet 4.6

 

[fdeck]

Here it is for fun. I always remind people that my design program has not been vetted by any real engineers, but is in pretty good agreement with WinISD if some assumptions are made.

Basic speaker modeling program

 

[ninjatanzen]

Here it is for fun. I always remind people that my design program has not been vetted by any real engineers, but is in pretty good agreement with WinISD if some assumptions are made.

Basic speaker modeling program

Based on your program, it looks like the speaker would handle about 25 watts at 60 Hz before reaching xmax. Definitely not ideal.

Using SBL program with a max overhead airline capacity of 22x14x9 inches, using 3/4" material, allows for roughly 1.08 cu/ft of useable airspace on that 10FE200 driver. It shows about 88 watts at 60 Hz before xmax is exceeded in this sealed design. Still not great, and it has -6.1 dB at 60 Hz. That cab with 13 ply baltic birch and that driver would also likely weigh around 30-35 lbs at minimum. I just don't see how the initial requirements are going to be possible. AI isn't magic (far from it), and OP's initial requirements are pretty substantial. Perhaps with some expensive driver in the hands of an actually qualified audio engineer something would be possible, BUT it would very likely be extremely cost prohibitive and have other tradeoffs (like efficiency).

For me, building a cab at home is about tinkering and having fun. When looking for a proper sounding/designed cab, I defer to the well educated.

 

[agedhorse]

From Claude Sonnet 4.6

Total bull-poopie.

First mistake is that the 10FE200 is not a Neo driver... and it goes downhill from there!

Not going to come close to their predicted performance specs, again another well presented, but steaming pile of...

 

[fdeck]

Indeed, Qtc can't be less than Qts.

Something I've experienced with the AI, even the latest versions, is that it hallucinates numbers that it can't copy from somewhere. For example you can compute Qtc from a widely accepted formula. The Google AI search result even gives that formula and it's correct. And the numbers for plugging into the formula are in the datasheet for the driver. But what AI can't do at this state of development is plug known numbers into a known formula and give you the result.

The whole thing is a disaster. Note that I'm not laying blame... I'll take it at face value that this is a non-judgmental exercise in seeing what the AI is capable of today.

 

[Jugghaid]

And now we are all even more aware of how crappy AI currently is.

 

[wimpy]

AI is a great tool and it will probably be the only thing left of humanity one day, but it still makes a lot of mistakes. But it has to be learned, even if it is sometimes sad and painful, like jumping from the Eiffel Tower in a coat with bigger sleeves, like wings..

 

[No such thing]

And now we are all even more aware of how crappy AI currently is.

Sure, and eventually it will be as commonplace in design as CNC is in manufacturing.
Probably sooner than later.

And just like with CNC, nobody will really notice anything different.

Life goes on

 

[agedhorse]

Sure, and eventually it will be as commonplace in design as CNC is in manufacturing.
Probably sooner than later.

And just like with CNC, nobody will really notice anything different.

Life goes on

CNC doesn't attempt to design a solution or fake a result.

 

[Just go for it]

AI

 

[Redbrangus]

Like any other tool, AI needs to be used properly. And like any other tool, there's something of learning curve to learn how to do that.

 

[MultiScaleMale]

AI will just string words or tokens (words split up) and given the previous context, which token should I pick next. (Super simplified) Repeat, repeat, repeat. This works great for things like creating text, creating code, etc. When you have it "create a cabinet" it's going to string together verbal descriptions to give a description of the cabinet. That includes both good and bad descriptions from online forums, vendor descriptions, probably even newsgroups. There is no insight into how a cabinet works other than the collective knowledge (and bogus uninformed data) that was put into text format.

I had Claude look at my tickets for the past six months and write a performance review for me within the guidelines of my company's guiding principles. Claude did a spectacular job with that.