Did you look at the videos? It does screw up calculations.
-
WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!
ChatGPT to calculate DIY speakers
- Thread starter sarumbear
- Start date
OP
- Thread Starter
- #102
Did you checked the calculations I posted? My post is only about using it for speaker calculations.
napfkuchen
Senior Member
I gave two of these problems with the exact same prompt to chatgpt and the results were correct for me.
But even if they are not and you kind of know that telling chatgpt it did s.th. wrong helps because it will apologize (get's annoying soon ...), "rethink" the prompt and most of the times put out the results I was looking for.
Just in case someone was wondering -
Are you Hal 9000?
No, I am not Hal 9000. I am ChatGPT, a language model developed by OpenAI. Hal 9000 is a fictional artificial intelligence character from the novel and movie "2001: A Space Odyssey" by Arthur C. Clarke and Stanley Kubrick. While Hal 9000 is a work of fiction, ChatGPT is a real technology that is designed to simulate human conversation and provide information and assistance to users.
We decided this based on experience using it and finding that it doesn't work. In my case, several hours of trying to get it to work.
It treats quantitative and qualitative information the same way, remember.
I don't know that anyone said it CAN'T work. It just doesn't work as it exists today.
ChatGPT today is not suitable for any work where correctness actually matters. Maybe the next version(s) will be better, I don't know. I think we all agree it's promising but not ready for prime-time, why is there any debate on this?
OP
- Thread Starter
- #106
Does that include speaker design by DIY enthusiast, which my posts was aimed at?
I don’t see how this is more useful than tools like Hornresp or WinISD? They actually show you the response, you can tweak it on the fly and see the result in real-time.
I don't think it's promising at all.
As I've said, the amount of computational power and investment in training make it seems it's more that what actually is. It's nothing smart, it's brute force. IMO it's a marketing device to convince people to pay to use OpenAI models, nothing more. Someone has decided to loan those people a lot of money under the promises that a lot of data scientists and researcher will start using OpenAI models and that will make huge profits.
The truth however is that those models aren't special at all, and that chat GPT gets its appearance from being hugely dopped, not because it's a clever model that changes anything. It's not an advance in the field of statistical learning, it's marketing, that's all.
OP
- Thread Starter
- #110
Both a kilogram of iron and a kilogram of hay weigh the same amount, which is one kilogram. The weight of an object is measured in units of mass such as kilograms, pounds, or grams. Therefore, in this case, both objects have the same weight because they have the same mass. It is important to note that the properties of the materials are different, and their densities, shapes, and sizes can affect how heavy they feel or appear to be when lifted.
Yes, for example, I tried to get it to design a crossover and it gave me conflicting instructions. As I've said before, I don't think it's useful unless you already know the answer or have calculated it with another method, in which case why bother?
I think this is a little bit like saying Jimi Hendrix was nothing special, he just played the same instrument louder and in a different way. In my tests with ChatGPT I've seen it do things with natural language processing that I didn't think I would see in this decade. Whether it's mere brute force or something more interesting, it can still take a very wide range of arbitrary inputs and do something useful (or sometimes, almost useful) with them. This is what I think is promising, despite the obvious flaws.
F1308
Major Contributor
- Joined
- May 24, 2020
- Messages
- 1,058
- Likes
- 910
Is that your answer or chat GPT's.
Because a friend of mine told me he got a very different inaccurate answer.
OP
- Thread Starter
- #113
I copy and pasted ChatGPT answer. Asked again the answer had different wording but the same correct one.
Both are equally heavy, as they both weigh one kilogram. The unit of measurement "kilogram" refers to a specific amount of mass, and it is a fundamental unit of the metric system used to measure weight and mass. Therefore, a kilogram of iron and a kilogram of hay weigh the same.
I asked bing a similar question and it said:
A pound of lead and a pound of feathers both weigh the same, about 16 ounces or 0.45 kilograms. Is there anything else you would like to know?
But note that it thinks a pound is only 'about' 16 ounces. So much for precision.
Bing? Chatgpd?
F1308
Major Contributor
- Joined
- May 24, 2020
- Messages
- 1,058
- Likes
- 910
This.
I wish others would understand this better. None of the current "AI" works the way folks say. That is projecting your learning model upon the AI of your choice. The models don't understand sentence structure anymore than paint colors... they are given the option to "say" anything they want trillions of times and given feedback that a give word order doesn't generate a high acceptance.
Think about that for a minute. AI models currently are trained by being told that their output is "pleasing", or "best", or "correct" by the users essentially.
They are no different than all social media.... they tell people what they want to hear. and if not, adjust outputs until people like what it has to say.
This.
I wish others would understand this better. None of the current "AI" works the way folks say. That is projecting your learning model upon the AI of your choice. The models don't understand sentence structure anymore than paint colors... they are given the option to "say" anything they want trillions of times and given feedback that a give word order doesn't generate a high acceptance.
Think about that for a minute. AI models currently are trained by being told that their output is "pleasing", or "best", or "correct" by the users essentially.
They are no different than all social media.... they tell people what they want to hear. and if not, adjust outputs until people like what it has to say.
OT
I wonder how they do that? Do they backpropagate the right answer 10000 times? So the wighting is heavy on it?
One thing ChatGPT does is just make things up - for example, if I ask about getting a refernence for protein X and disease Y, it just comes up with a completely bogus (but real-looking) reference. If I then challenge it about this and tell it to try again, it comes up with a real reference - it's a bit like a student who thinks their questioner won't bother looking to see if they have plagiarized something.
So, essentially a child, IMHO
So, essentially a child, IMHO
As for Alexa, it would have comparable issues, based on this from recent ZDNET article…
…(Amazon CEO Andy) Jassy acknowledges that Alexa was powered by a large language model but aims to equip the virtual assistant with one that can execute more complex tasks…
As with ChatGPT, if I did not find its mistake before building would be wasting my precious DIY time and money!
