- Joined
- Aug 16, 2021
- Messages
- 120
- Likes
- 188
Doesn't matter if I agree or not, this is what ChatGPT had to say on the issue:
Oh, I got back to youTry asking if it affects the sound, rather than how it affects the sound, and get back to us.
You could also ask why RCA cables don't affect the sound.
And while you're at it, ask it how we know that the Earth is flat.
Well, we're already going from them definitely sounding different, to "unlikely to be significant in most cases".
Yes, I understand how transformer networks work. ChatGPT will not 'hone' in on 'what I want to hear, it will simply respond according to the data it was trained on. So, taking your example, me wanting to hear that 'The Earth is flat' will not force or 'hone' GPT to give me an answer other than 'the earth is round'. So yes - unlikely to be significant in most cases means significant in some cases, which is pretty much saying not much at all. I do find it interesting though that GPT follows the 'consensus' opinion, while the 'cables don't have a sound' data - which surely it must have been trained on - seems to be weighted less.Well, we're already going from them definitely sounding different, to "unlikely to be significant in most cases".
My point isn't that I or ChatGPT are "wrong", but that the answer that it gives is highly dependent on the questions and what it has "learnt". Because it also tries to give a balanced answer, the way you read the answer can be just as important. See how I have ignored that the second answer still ends with "listen carefully to the differences in sound"?
It's algorithmically playing the read-the-customer game as well, because your next question will have it hone in on what you want to hear in most cases, just as a medium will make educated guesses until you say "yes, that's my old aunt" or whoever.
How do you know that? Do you have access to the training dataset? I'd assume the exact opposite - that it was trained on the subjective mumbo jumbo which dominates in terms of volume, and that it doesn't read or interpret the AP plots which substantiate the scientific POV.I do find it interesting though that GPT follows the 'consensus' opinion, while the 'cables don't have a sound' data - which surely it must have been trained on - seems to be weighted less.
....it will simply respond according to the data it was trained on....
How do you know that? Do you have access to the training dataset? I'd assume the exact opposite - that it was trained on the subjective mumbo jumbo which dominates in terms of volume, and that it doesn't read or interpret the AP plots which substantiate the scientific POV.
This is still worthless lol. Ask it for a complete list of sources for the answer you pasted in the OP.
I agree. "Here is the bullshit produced by the bullshit generator." At least put in the effort to write your own.omg it just regurgitates whatever BS is most common on the Internet. I wish mods would take a look all this spamming with ChatGPT copypasta
The only real use for ChatGPT is to send texts back to your partner...Exactly - garbage in, garbage out.
People - please stop trying to use Chat GPT as an encylopedia - it isn't.
Then we are down to the capability of the curators. Do you think these curators are expert in all topics, so they can assure that only factual content gets put into the training set? Over how many billions of data points? Especailly with the variaty of sources including web pages - which as we know, in large part have little to no validation.