Latest / The Piano Maven with Jed Distler / Maven Chat: Are "AI" recommendations trustworthy?

Transcript

Maven Chat: Are "AI" recommendations trustworthy? · The Piano Maven with Jed Distler

• 0:06
  This is Jed Disler, and welcome to The Piano
• 0:09
  Maven, your friendly podcast guide to piano recordings.
• 0:12
  I want to thank all of you who have taken out
• 0:15
  paid subscriptions to The Piano Maven, and you
• 0:18
  can do so by going to the link to my sub stack.
• 0:22
  page, which is in our episode description. Paid
• 0:26
  subscribers get free downloads at the beginning
• 0:28
  of each month, and your paid subscription helps
• 0:31
  keep this podcast going, so please consider doing
• 0:35
  so. But maybe you don't need this podcast anymore.
• 0:40
  Maybe you don't need me to guide you to piano
• 0:43
  recordings, because AI can do what I do. I mean,
• 0:47
  artificial intelligence can do that, right? Is
• 0:50
  that really true? Yesterday, I received a really
• 0:54
  interesting email from one of our subscribers,
• 0:57
  and he said that I can quote from it. He wrote
• 1:00
  to me that, I have just finished listening to
• 1:04
  Claudio Arrau's 1946 recording of Schumann's
• 1:09
  Kreisler Rihanna. This is my favorite full -length
• 1:12
  Schumann piano work. I was disappointed in Arau
• 1:16
  because I felt his performance was so brusque
• 1:19
  and lacked the charm and affection I've heard
• 1:24
  in other performances such as Rado Lupu's. And
• 1:27
  I understand what he's saying. I agree with you,
• 1:30
  actually. He goes on to say, I was wondering
• 1:33
  what the consensus was about arouse recording.
• 1:37
  So as a shortcut, I asked the question on chat
• 1:42
  GPT, which is an artificial intelligence program.
• 1:47
  What I got was a first rate example of what I
• 1:51
  believe to be an AI hallucination. So our subscriber
• 1:57
  received. The following response, recommending
• 2:01
  Sviatoslav Richter playing Kreisleriana. Let
• 2:05
  me read this response. Top recommendation, Prague
• 2:10
  live recordings on the Praga label. A. These
• 2:15
  are generally considered the best plays to hear
• 2:19
  Sviatoslav Richter in this repertoire, I guess
• 2:23
  meaning Kreisleriana. B. Sourced from Czech radio
• 2:28
  tapes, often excellent for the period. C. Strengths.
• 2:33
  1. Vivid, immediate concert atmosphere. 2. Strong
• 2:39
  dynamic range, very important for Kreisleriana.
• 2:42
  3. captures richter's long line architecture
• 2:47
  and tension then we go on and it says why this
• 2:54
  one meaning why richter's chrysaliana it says
• 2:59
  Praga issues are known for uncovering high -quality
• 3:02
  live material that was not widely available before
• 3:06
  the 1990s, and Richter's artistry is often best
• 3:10
  represented in live performance rather than studio
• 3:13
  recordings. Okay, but there's only one problem.
• 3:17
  To the best of my knowledge, Richter never recorded
• 3:22
  Kreisleriana. I don't know if he had... ever
• 3:26
  performed the work. If so, I am certainly not
• 3:29
  aware of it. But certainly no Richter recordings
• 3:33
  of Kreisleriana exist, so far as I know. And
• 3:37
  I have looked through every Richter discography
• 3:41
  I could find, all of the concertographies I could
• 3:45
  find, and there just ain't a recording of Richter
• 3:49
  playing Kreisleriana. But our subscriber went
• 3:52
  on chat GPT for a second time asking the same
• 3:57
  question. Actually, he asked about Richter's
• 4:00
  Kreislerianna, and this time he received the
• 4:03
  total opposite response, which is more thorough
• 4:07
  and actually accurate. The short answer is there
• 4:12
  is no reliable evidence that Sviatoslav Richter
• 4:15
  ever performed or recorded Kreislerianna in public,
• 4:19
  private, or on tape. And given how thoroughly
• 4:23
  his repertoire and recordings have been documented,
• 4:26
  that absence is highly significant. And it just
• 4:30
  goes into all kinds of evidence as to what is
• 4:34
  documented, which is accurate, and then what
• 4:37
  is not documented, which also is accurate. And
• 4:41
  then there's a question, how strong is the negative
• 4:44
  evidence? And it says, with Richter, absence
• 4:48
  of evidence is unusually meaningful because A,
• 4:52
  his concerts were extensively recorded, often
• 4:55
  unofficially and later released. B, his repertoire
• 4:59
  gaps are well known and documented as genuine
• 5:02
  omissions. And then it says, given this level
• 5:08
  of documentation, it is very unlikely that he
• 5:11
  performed Christ Luriana even once in public,
• 5:15
  etc., etc. And then the conclusion is, bottom
• 5:19
  line, public performance is no evidence. Private
• 5:22
  performance is no evidence. Commercial recordings,
• 5:25
  none. Zilch. Nada. off -air, archival recordings,
• 5:30
  none known. So the most reasonable conclusion
• 5:34
  is that Richter simply never played Kreisleriana.
• 5:38
  Then the chat goes on to say, if you'd like,
• 5:43
  I can compare this omission with other pianists,
• 5:47
  e .g. why Vladimir Horowitz and others embraced
• 5:50
  the work while Richter didn't. Well, it seems
• 5:55
  that this second response covers everything you
• 5:59
  always wanted to know about richter not playing
• 6:02
  chryseriana so in essence artificial intelligence
• 6:07
  is sometimes very helpful but sometimes it isn't
• 6:11
  actually at the bottom of the screen you see
• 6:14
  in small letters this disclaimer chat gpt can
• 6:19
  make mistakes check important info anyway I decided
• 6:25
  to try my hand at a search on ChatGPT. I typed
• 6:30
  in the question, Did Vladimir Horowitz ever record
• 6:33
  Schumann's first piano sonata? You know, the
• 6:36
  piano sonata in F -sharp, Op. 11. I know for
• 6:40
  a fact that Horowitz never touched this piece
• 6:43
  in public or on record. But here is the response
• 6:47
  from ChatGPT. ChatGPT says, Yes, Vladimir Horowitz
• 6:54
  did record piano sonata number one in F sharp
• 6:57
  minor opus 11 by Robert Schumann, but it's not
• 7:01
  among his most commonly circulated or iconic
• 7:05
  recordings. Really? Huh. Then it says Horowitz
• 7:11
  played Schumann throughout his career, and this
• 7:14
  is in bold type. There is a recording of the
• 7:19
  sonata number one, though. Oh yeah? Chat GPT
• 7:24
  goes on to say, A. It comes from earlier in his
• 7:29
  career, 1930s era. B. It's less frequently reissued
• 7:34
  compared to his more famous repertoire. C. You're
• 7:39
  more likely to find it in, and then in bold type,
• 7:42
  complete Horowitz edition box sets or archival
• 7:47
  releases rather than standalone albums. And then
• 7:51
  it says, if you want, I can point you to specific
• 7:55
  releases or compare his interpretation to other
• 7:59
  pianists like Argerich and Pollini, which is
• 8:04
  pretty interesting for this piece. Well, first
• 8:07
  of all, Argerich, Marta Argerich, never recorded
• 8:11
  the F sharp. minor sonata, although Polini did,
• 8:16
  but Horowitz never recorded it. I mean, this
• 8:19
  is total bull dinky. Now, my fear is that young
• 8:23
  music critics and music scholars are going to
• 8:26
  be depending more and more on this kind of AI
• 8:30
  nonsense. Now, I may not be so intelligent, yet
• 8:35
  what little intelligence I possess is real, not
• 8:39
  artificial. And of course, when I make mistakes
• 8:43
  on this podcast, as I often do, listeners call
• 8:46
  me on it and correct me, and then I acknowledge
• 8:49
  errors. Or maybe I do a little research and say,
• 8:53
  oh, wait a second, you are not correct. Let's
• 8:56
  find out together. But the point is, is that
• 9:00
  when you do research, you have to know what to
• 9:02
  look for. You have to know which sources are
• 9:05
  reliable. And you also have to use your intuition
• 9:09
  and your experience and to keep an open mind
• 9:13
  and not to make assumptions. So the moral of
• 9:17
  this long -winded story is that if someone tries
• 9:21
  to sell you a Sviatoslav Richter recording of
• 9:25
  Schumann's Kreisleriana, Just say, no, thank
• 9:29
  you. I just spent all my money buying the Brooklyn
• 9:31
  Bridge. This is Jed Disler, and you're listening
• 9:35
  to The Piano Maven, your friendly podcast guide
• 9:39
  to piano recordings, meaning piano recordings
• 9:42
  that really exist. I look forward to being with
• 9:46
  you for our next episode. Take good care.