Just made a custom GPT, feeding it multiple books and white papers, trained it on the data. Curious to know what you guys think. You can give it a try here:
https://chat.openai.com/g/g-CLP2DodWv-speaker-design-guru
https://chat.openai.com/g/g-CLP2DodWv-speaker-design-guru
Cool - I will try it out!
Now, could be you so kind to integrate GPT-4 or Copilot into VituixCAD, Hornresp, Ath4, AKABAC/ABEC, and Acourate, please?
What would be cool though is if you could scrape this whole forum, and the absolute wealth of information contained herein but so hopelessly diffused and time consuming to find, and train a custom GPT on that. It would need to be "intelligent" enough to sort out the wheat from the chaff though, and not sure we are there yet.
- "Alexa, generate G-code for a MEH for me, based on x and y dispersion, z Hz cut-off, and pick the best drivers from manufacturers a, b an c based on actual measurements found online."
Now, could be you so kind to integrate GPT-4 or Copilot into VituixCAD, Hornresp, Ath4, AKABAC/ABEC, and Acourate, please?
What would be cool though is if you could scrape this whole forum, and the absolute wealth of information contained herein but so hopelessly diffused and time consuming to find, and train a custom GPT on that. It would need to be "intelligent" enough to sort out the wheat from the chaff though, and not sure we are there yet.
- "Alexa, generate G-code for a MEH for me, based on x and y dispersion, z Hz cut-off, and pick the best drivers from manufacturers a, b an c based on actual measurements found online."
Has anyone got anywhere using AI to guide technical decisions? I once tried to use it to help sort out some damping models (for speakers) but it got a fair amount wrong that I recognized as wrong and likely some that I didn't. Attempts to fix it by adding more questions/statements seemed to make matters worse. Now I didn't know what I was doing but the unreliability put me off. It would be good to hear from anyone that has found it useful and how they addressed the unreliability which I guess is unavoidable to some extent.
It's REALLY bad at math, as in like extremely bad, as in any person who just started a course in algebra is better.
I was trying to use it for a simple resistor divider network, just because I was lazy one day.
It could just not give the right values, even when I corrected it like 5 times.
I did something similar with cascading Q-factors.
It keep saying that a Q of 0.5 was Butterworth (or the other way around, can't remember)
I mostly use it for text base stuff.
It it really good in like spitting out ideas for like synonyms, or coming up with names or so.
It's also quite decent with simple bulk programming code language problems.
Like sure, you still need to fix the errors, but at least I don't have to do all the monkey work anymore
So yeah, my experience is the same like yours, including the really bad vibe with unreliability, yet "sounding" very convincing.
But the thing that put me off the most, is that it just totally refuses to give ANY kind of resource.
Not even like a a Wikipedia page, a good book in the field or whatsoever.
That is for any kind of scientific subject a direct no-no obviously.
Which is ironic since it's written by people with such backgrounds, so they even should know that extremely well.