I worry about this. Dumping data and getting a report on which actions are based prolly isn’t giving real data.
I say that because I have Claude review code and then review the review and it will point out all the things it got wrong the first time. And I end up having to review it myself. Don’t get me wrong, I’m burning tokens all day and love the legit productivity gains but I don’t see how this replaces everybody in tech for a while.
when gmail launched in 2004, they did indeed target ads based on email content. People did not like it. Apparently in 2017 they ended "the practice." But now, about 10 years later, it's back?
It’s just cheaper. It has nothing to do with it being more listenable. Same with auto tune. It’s cheaper to have some mediocre singer do the track and fix it later. You don’t have to pay the great singer. And if you do get a great singer, you don’t keep them in the studio for as long. It’s just a cost savings. It doesn’t improve the music.
How about a situation where you're recording a song with your buds/bandmates and the drummer played a few notes out-of-time or singer didn't quite sing that long note in tune. You're not going to hire anybody to replace what they did. You can have your buds come back and try and try until they get something to your standard or tell them that they're bad musicians and need to study and practice for more years and years before they dare want to record a song again (purist). Or You can just live with the bad ("shitty is pretty"). Or you could hire someone to come in and replace their parts. Or you could just take a few minutes to fix it in cubase.
To me, the central question is: is this about the composition or the performance? Personally, I spent my university time getting a degree in music composition, so getting the point of the composition across is my priority. I'll auto-tune and quantize whatever if the composition comes across better. But I could understand if other people feel differently (they are a purist or very performance-oriented).
I mean, there's a difference between nudging a few notes right and just running quantize over the entire recording.
> so getting the point of the composition across is my priority. I'll auto-tune and quantize whatever if the composition comes across better.
Which also means you might even move things out of sync/tune if the composition asked for it, right? The point mentioned here is just the trend of lazy fully gridded mixes where the entire song is the exact bpm all the way through and nothing strays off the grid.
It's cheaper to do what? The thing they're getting for less money is precisely the improved quality or listenability. Their definitions of quality and good enough might be fucked up, but the intent and purpose is clearly to improve it up to some minimal level of quality, otherwise how are they judging a mediocre vs. great singer? How do they know when the great singer is done singing (sang a good enough take), or when they've tweaked the mediocre vocal track enough? The technological interventions change the ratio of cost vs. quality, they don't make quality irrelevant. If they did, you could just record the crappy singer as-is and save even more money.
This post's point is that musical "quality" is not a unidimensional feature and that mechanically optimizing for these "quality" metrics can kill something interesting in music.
Yes I get it, and I'm not replying to the post. I'm replying to the comment that said it was all about saving money and "has nothing to do with it being more listenable." Clearly that can't be true; these producers are trying for some standard of listenability (which I already said could be wrongheaded, to your point). If they weren't, they would not only save money by hiring mediocre talent, they would also save even more money by not doing any post-production interventions at all.
> It’s cheaper to have some mediocre singer do the track and fix it later. You don’t have to pay the great singer. And if you do get a great singer, you don’t keep them in the studio for as long. It’s just a cost savings.
I think you're on to something, but it isn't quite what you think it is.
Corrections are largely a cost issue. But not because singers are so expensive.
The costs being saved are related to the rarity of finding someone who is a great singer at the same time that they're also a great dancer and stunningly attractive. It's much easier to find someone who's an outlier on one dimension than someone who's an outlier on two. If you want someone who is extreme on three dimensions (that you specify in advance), you probably won't be able to find that person at all.
Corrections in production mean that you can achieve something like this:
- Beauty: extreme natural talent, plus help from consultants, stylists, makeup artists, etc
- Dance: about what an average person can achieve through intensive training
- Vocals: extreme machine-supplied talent
at low cost. There are other possible approaches. You could look for a great singer who's willing to have tons of plastic surgery. You could look for a great dancer who's willing to have their singing and physical appearance 'corrected' for a trifecta. I don't actually think either of those would make for a cultural improvement - to the extent that we aren't already doing them - compared to what we have now.
I switched to NA beer so that when I did decide to “have a few” it would be a rare occasion. Then I realized that NA beer was doing it for me and I haven’t gone back. It’s not for everyone but it worked for me.
Ah civ 1. Been chasing that high ever since. I talked my parents into buying a 486sx “for school” but I made sure it met the min reqs for civ. I put that packard bell through so much.
I've found sbx to be very helpful. really like the sentinel value wrapper they have going so you can add secrets but the model can't see them. Outbound calls get looked up by sentinel value and the real one goes out to whatever API you're auth'ing too. There's other features but just having claude run in a sandbox and easily see what it does and doesn't have access to has been great.
I just wish sharing whitelisted domains between sandboxes were easier or more intuitive. Last time I tried doing this with sbx I was handling their yaml files by hand.
Being able to drag a control onto the form and then double click it to get to the event code is just magical. I remember very well clicking around and building things for several hours before reaching for a docs back in the day.
I say that because I have Claude review code and then review the review and it will point out all the things it got wrong the first time. And I end up having to review it myself. Don’t get me wrong, I’m burning tokens all day and love the legit productivity gains but I don’t see how this replaces everybody in tech for a while.
reply