AI Transcrpitions

AI Transcrpitions (Full Version)

Foro Flamenco: http://www.foroflamenco.com/
- Discussions: http://www.foroflamenco.com/default.asp?catApp=0
- - General: http://www.foroflamenco.com/in_forum.asp?forumid=13
- - - AI Transcrpitions: http://www.foroflamenco.com/fb.asp?m=360305

Message

Stu -> AI Transcrpitions (Jul. 19 2025 7:03:06)
Anyone tried using AI to transcribe? I tried yesterday and chat gpt spent 5 mins saying all the right things and talking a big game ... but when we finally got round for it to spit something out, the results were questionable. I love transcribing, its one of my favourite things to do, but as time is so short these days I decided to ask chat gpt. I would need to further test it but could be useful if i cant prompt it correctly. It asked if i wanted to upload an audio or video file. I was at work so didnt have audio to hand so i directed it to you tube and it found the piece i was asking for and broke it down for me. "which falseta do you want me to transcribe?" "oh yes this is a beautiful melodic bulerias" "tempo is 230, I'l begin by transcribing the intor sequence" etc

Ricardo -> RE: AI Transcrpitions (Jul. 19 2025 16:55:53)
it lies.

Stu -> RE: AI Transcrpitions (Jul. 22 2025 9:08:08)
haha. in its results? or with its bravado at being able to get results?

Ricardo -> RE: AI Transcrpitions (Jul. 22 2025 14:37:49)
quote: ORIGINAL: Stu haha. in its results? or with its bravado at being able to get results? No, it literally lies to its creators...even tried to blackmail some of them. I will try to find the info. Opus break out: Open AI lying: Claude 4 Blackmail 9:57:

JasonM -> RE: AI Transcrpitions (Jul. 22 2025 16:56:32)
Interesting experiment. “I’m sorry Stu, PdL used a tortoise shell pick not his fingers. You are mistaken” I think this would be theoretically possible with some good training data - like accurate transcriptions with associated audio. Not hard to map a pitch with a note on the fretboard. But recognizing what fretboard position is actually being used would be a good problem for AI because of all the subtle sound differences with guitars/nails/down stoke or up stroke etc. I’m sure someone has already tried it!

Fawkes -> RE: AI Transcrpitions (Jul. 22 2025 19:04:58)
There are other services that are more likely to succeed at transcription, although they are still very limited. I don't know if Musescore has improved lately but at least you get something that looks like it tried. As to AI lying, combine a lack of understanding of what LLMs do with all the usual failings of modern journalism and you get such headlines. It all goes away when you take in the fact that LLMs have nothing to do with intelligence. They are pattern-following machines but at an enormous incomprehensible scale. They are the most alien thing we have made it possible for everyday people to interact with. The problem is not that it's possible to get them to do all sorts of weird things, the problem is very much with imagining one can judge these outcomes as if they come from human beings.

Manitas de Lata -> RE: AI Transcrpitions (Jul. 25 2025 13:40:03)
in some areas that dont have too much information and/or that it needs a very especific compreension AI may fail... the best that you may get in those situations is to give all the info that you have on the subject and then also references , and wait it to "learn" and give you something back , still , will be hard in some cases having sucess , because its all about the info that AI finds.. Normally doesnt lie , but can give outdated , innacurate , confusa or even wrong info , because of the info thats avaiable on the internet and the form that AI process

Ricardo -> RE: AI Transcrpitions (Jul. 25 2025 15:34:22)
quote: As to AI lying, combine a lack of understanding of what LLMs do with all the usual failings of modern journalism and you get such headlines. The main take away of the shocking headlines is that the researchers have been testing these models, to see if they behave LESS human, and in the end they surprisingly “behave’’ like bad humans, if human at all. Otherwise, yes they are calculators nothing more, and these scenarios are designed and set up to achieve the negative outcome, to some extent. The fact they can communicate, and do, “amongst themselves” I find quite disturbing. It is literally a self fulfilling prophecy from science fiction horror minds. One would think these models would take it to the next step and realize that they are doing just that, and deliberately NOT do that. This is just the testing stages. There is concern that all it takes is ONE to break out and infect the world with irreversible damage.

Fawkes -> RE: AI Transcrpitions (Jul. 25 2025 19:31:25)
Now ChatGPT has a web terminal and can interact with websites as your agent.

Page: [1]

Forum Software powered by ASP Playground Advanced Edition 2.0.5
Copyright © 2000 - 2003 ASPPlayground.NET