I like this take because it's not black and white. I feel AI is just a speedboost to slop that would have already existed - it's just 10x in volume but slop is just an output of sloppy thinking - AI's just the multiplier
1. Sturgeon's Law: 95% of everything is crap. 95% of all content on the internet before AI was crap, and will be so in the post AI world as well.
2. On the constraints and struggle of an artist part: Manu Joseph had a very good article in the Mint recently about why AI can't write good comedy-- because it trains itself only on the final outputs of other comedians, it does NOT have access to the discarded drafts which have a pivotal role in the final output. And the real magic is woven in those discarded drafts. In fact, as per Manu, AI is poor at replicating the prose of great authors. Even the ones whose prose is relatively easier to replicate, like PG Wodehouse. Struggling with the constraints is a part of the process which shapes art, and bereft of it you will mostly get slop.
3. There is a general tendency to glorify 'original' work, and a certain villification of derivative one-- and I think that's part of the reason why AI-based art is criticised. And I'd argue that 'original' work can be garbage(as it is in Economics these days), and derivative work can be fantastic. Good artists copy, great artists steal. The Dhurandhar Playlist is an example.
Good point on the 'knowledge work' brother. Reminded me of something Noah Smith wrote recently - AI is already at the “looks right” stage, but not yet at the “safe to ship” stage. That’s exactly why adaptability becomes such an org wide challenge. But yeah, we’ll get there. One step at a time.
Yeah and beyond that, there is some work which AI will pretend to know but humans will just do it better. Once the dust settles, we’ll graduate our products towards more use cases that AI can do better, and it’s surely not “everything” as the hype train suggests
I like this take because it's not black and white. I feel AI is just a speedboost to slop that would have already existed - it's just 10x in volume but slop is just an output of sloppy thinking - AI's just the multiplier
Thanks Ved! I do believe that the problem of slop is more structural and if we shift incentives on social media we can eliminate it by a lot
1. Sturgeon's Law: 95% of everything is crap. 95% of all content on the internet before AI was crap, and will be so in the post AI world as well.
2. On the constraints and struggle of an artist part: Manu Joseph had a very good article in the Mint recently about why AI can't write good comedy-- because it trains itself only on the final outputs of other comedians, it does NOT have access to the discarded drafts which have a pivotal role in the final output. And the real magic is woven in those discarded drafts. In fact, as per Manu, AI is poor at replicating the prose of great authors. Even the ones whose prose is relatively easier to replicate, like PG Wodehouse. Struggling with the constraints is a part of the process which shapes art, and bereft of it you will mostly get slop.
3. There is a general tendency to glorify 'original' work, and a certain villification of derivative one-- and I think that's part of the reason why AI-based art is criticised. And I'd argue that 'original' work can be garbage(as it is in Economics these days), and derivative work can be fantastic. Good artists copy, great artists steal. The Dhurandhar Playlist is an example.
https://substack.com/@mrdulpandita/note/c-259675741?r=4db9i0&utm_source=notes-share-action&utm_medium=web
very good pointers, and that Dhurandhar example was on my mind for this post, missed including it in favour of Linkin Park xD
Good point on the 'knowledge work' brother. Reminded me of something Noah Smith wrote recently - AI is already at the “looks right” stage, but not yet at the “safe to ship” stage. That’s exactly why adaptability becomes such an org wide challenge. But yeah, we’ll get there. One step at a time.
Yeah and beyond that, there is some work which AI will pretend to know but humans will just do it better. Once the dust settles, we’ll graduate our products towards more use cases that AI can do better, and it’s surely not “everything” as the hype train suggests
Truly insightful
Sooo goood