Increasingly, people are using AI to answer questions using tools like ChatGPT, Gemini, Grok, etc. And, increasingly, we are being pushed into getting these answers by default through the search engines we use every day.
In my experience, these tools cosplay at being experts: they confidently state an answer and, despite often being very wrong, may even rigorously defend that answer when challenged.
It’s for this reason I’m, personally, scornful of people who cut/paste answers from such tools into forums like this.
In the Ardour (a popular Open Source DAW) forums recently, there was a discussion on the origins and pronunciation of “Ardour”. The correct answer is in the manual:
But one of the forum members, as an experiment, asked ChatGPT, which confidently gave this, entirely false, answer:
Very interesting Keith, well posted.
A funny thing about AI, chatGPT can now be purchased, and no one it appears is interested.
As one sage remarked along the lines of ‘if the people coming up with this stuff make it so intelligent, why do they not use it to make even more money?’
But there are LOTS of tools out there with other limited uses of AI that are exceptionally handy. Some of them I use on an almost daily basis. I definitely don’t group LLM chatbots in with this.
One forum I’m on installed its own AI chatbot some time ago (based on Google’s chatbot). It’s definitely hot garbage and when it was implemented, a lot of forum members challenged it hard to demonstrate that it’s hot garbage. Of course, the forum owners don’t care. They thought that programming it for sarcasm and jokes would fix that. It got worse.
Is it a ban on posts entirely generated by machines/AI (eg an answer to a question/a full song), or a ban on posts that contain AI to some extent
re: the pronunciation of Ardour- Years later, another interpretation of “Ardour” appeared, this time based on listening to non-native English speakers attempt to pronounce the word. Rather than ‘Ardour’, it became ‘Our DAW’
I am not aware of a single group of non-native English speakers who do not pronounce the ‘R’ in some fashion… only folk from England have difficulty with this
@Majik I am starting to see much more generated texts and pictures… sad part about this is that most people act like they are smartest ones and no one would notice.
This is the time where smart people use this tool for getting smarter and dumb people getting… you know.
The content that is prohibited is “machine generated”. Note that this part of the T&Cs existed before the widespread availability of AI and was, I believe, to prohibit the use of “bots”.
That definition could cover a lot of bases, but the moderators have made it clear that that posting of AI generated answers is not welcome.
Of course, Ardour, being a DAW, encourages community members to post links to content that has been produced with Ardour. There is the potential for such content to include AI generated materials.
I’m not sure that has been broached yet, but I suspect it will be “it depends”.
In all these things, human moderators make judgements on these things and make decisions accordingly. And, as on this forum, I trust them to make rational and sensible one.
Good luck! My grandmother was from Denmark (Ålborg). She taught me a few Danish words when I was little. Some I could pronounce, but many I simply could not do the back of the throat sounds that the words required.
Interesting thread and I’ve been wondering about that myself.
I have been experimenting with AI assisted coding and it does some odd / wrong things very confidently… just as confidently as the good things it does. Asking it to explain its thinking and cite its sources and having a dialogue seems to help.
So I decided to try Copilot’s AI to ask questions about music. And to my surprise, so far it’s twice cited Justin’s lessons and discussions in this very community as a source for its answers. In those cases (reactive listening, a theory question), its answers were correct. But there were times when it was very wrong but very confident about being correct. For example it claimed it could give me a structured plan to get better at blues soloing with dynamics and feel and emotion. While the structure made some sense the little etudes / tabs it generated to work on a specific thing were wrong most of the time - strings and frets labeled incorrectly, notes not from the scale etc.
The think deeper mode does seem to generate better answers. For example:
Which chatbot have you managed to convince to cite its sources? The one on the other forum that I’ve dressed down absolutely refuses to cite sources (except when it does, and when I point that out, it gives some BS answer about why it cited that particular source, but nothing else).
It’s the Microsoft CoPilot app for iPhone. I asked it to name the models it was using and it said “I’m sorry, but I don’t have details on the specific model names behind this assistant. I’m built on Microsoft’s own proprietary large-language models and not affiliated with Claude, ChatGPT, or other third-party systems.“
Some of the AI generated text you can read online is just silly. My two hobbies are motorbikes and playing guitars. You can read more and more “reviews” of new models of motorbikes and guitars that are just made up based on the specs from the manufacturers press release. Sometimes the flaws in the so-called reviews are hilarious.
I was reading a review of an epiphone guitar that had 2 pickups; a probucker 2 and a probucker 3. The review said something like “ this guitar has a great sound because it’s got 3 pickups”
Another review of a guitar said that it’s a step up in quality because this guitar has a straight neck.
Even better for motorbikes. I’ve read that one particular model is faster because it has footpegs.
I also read a review saying a bike is speedy because it’s red. In the manufacturers press release, the name of one of the paint colours was « Speed Red ».
