It's extraordinary to me the relentless lack of critical thinking and understanding of how these models work that's being exhibited upon the wider Internet, in relation to generative AI. Always appreciate your thoughts Johan.
I know Glaze and Nightshade exist for protection of human- made images, I hope something can be created to also protect music and writing from being illegally scraped too.
No worries, Johan! One of the best AI discussions I've had - although I did have a pretty deep one with "Sky" on Open AI (before they pulled her!) - it went 12 minutes, and she asked really pertinent and intelligent questions. One day we'll grab a bite in person! Cheers -
Hi Johan - Just thought I'd include our discussion on LinkedIn (which you used BTW without attribution, license or compensation - all that you say you stand against in our discussion) that formed the basis for this post:
Steve Stewart
CEO & Co-Founder, Advisor, Former Stone Temple Pilots Manager
1w
Anyone open to a thoughtful, in-depth and intellectual discussion around the following question:
As someone who has learned to play a number of instruments over the years, I have spent many thousands of hours training my brain on copyrighted songs, copyrighted sheet music, and copyrighted music videos to learn musical skills from other musicians. Everything I play or write is built directly on the training of my brain on the copyrighted work of others. I would say this is true of every other human that has ever learned how to play an instrument or write a song, book or script. I have not used anyone's actual recordings in my work, nor have I copied their exact words or lyrics. I have derived financial benefit from it, and have not included any of these pervious influences in any music or writing credits I have issued, and have not made any payments to any of them, outside of legally purchasing their copyrighted materials (records, sheet music, streams, downloads, etc.). Other than being exponentially faster, broader and more accurate, how is AI doing anything different than this?
I understand NIL concerns, but musical styles, basic chordings and progressions cannot be protected as far as I know.
…see more
Johan Cedmar-BrandstedtView Johan Cedmar-Brandstedt’s profile
Steve Stewart the difference is, you’re a person.
It’s honestly a very weird framing to compare mass statistical processing of billions of specific song recordings on a data center for hundreds of thousands of dollars to the embodied, encultured, reciprocal, dynamic, living process of human learning that takes in a myriad of different sources.
In a legal sense, you are a person consciously composing each element of the song. If you emulate a pre existing song, you most likely do so consciusly. If you want to borrow a sample, you clear it. If you do a cover, you credit the original songwriter. And when you perform it, it expresses your personality: your limitations and idiosyncrasies as a performer, your choice of instrument, acoustics, effects. Your subjective interpretation of the emotional resonance and meaning of the song.
Whereas model is nothing *but* samples. A statistic compilation of underlying works, from which you can extract derivatives.
I don’t really see any similarity at all. Not ontologically, not technically, not legally, not morally.
Thanks, Steve! I did paraphrase your parts, and I did tell you about this one. Wasn’t sure you would appreciate the attribution, but now that I know you want it, I’ll of course include it. Cheers!
An alternative to Adobe is attracting some attention - Affinity by Serif (now owned by Canva). Currently 50% off for a perpetual multi-platform, 3-tool bundle: https://affinity.serif.com/en-us/affinity-pricing/
Hi Johan :) I read through Affinity’s T&Cs and didn’t see any red flags. I haven’t checked Canva’s yet (don’t use it). New corporate parents can definitely drive changes. What makes you most uneasy about Canva?
Ok, so I had a hunch I had heard something bad, and sure enough: Canva provide Stable Diffusion image generation in their products. Stable Diffusion is a research model not licensed for commercial use. They also partnered with OpenAI. So even if their products don’t seem that data greedy today, the parent company is rotten.
It's extraordinary to me the relentless lack of critical thinking and understanding of how these models work that's being exhibited upon the wider Internet, in relation to generative AI. Always appreciate your thoughts Johan.
I know Glaze and Nightshade exist for protection of human- made images, I hope something can be created to also protect music and writing from being illegally scraped too.
No worries, Johan! One of the best AI discussions I've had - although I did have a pretty deep one with "Sky" on Open AI (before they pulled her!) - it went 12 minutes, and she asked really pertinent and intelligent questions. One day we'll grab a bite in person! Cheers -
Hi Johan - Just thought I'd include our discussion on LinkedIn (which you used BTW without attribution, license or compensation - all that you say you stand against in our discussion) that formed the basis for this post:
Steve Stewart
CEO & Co-Founder, Advisor, Former Stone Temple Pilots Manager
1w
Anyone open to a thoughtful, in-depth and intellectual discussion around the following question:
As someone who has learned to play a number of instruments over the years, I have spent many thousands of hours training my brain on copyrighted songs, copyrighted sheet music, and copyrighted music videos to learn musical skills from other musicians. Everything I play or write is built directly on the training of my brain on the copyrighted work of others. I would say this is true of every other human that has ever learned how to play an instrument or write a song, book or script. I have not used anyone's actual recordings in my work, nor have I copied their exact words or lyrics. I have derived financial benefit from it, and have not included any of these pervious influences in any music or writing credits I have issued, and have not made any payments to any of them, outside of legally purchasing their copyrighted materials (records, sheet music, streams, downloads, etc.). Other than being exponentially faster, broader and more accurate, how is AI doing anything different than this?
I understand NIL concerns, but musical styles, basic chordings and progressions cannot be protected as far as I know.
…see more
Johan Cedmar-BrandstedtView Johan Cedmar-Brandstedt’s profile
Steve Stewart the difference is, you’re a person.
It’s honestly a very weird framing to compare mass statistical processing of billions of specific song recordings on a data center for hundreds of thousands of dollars to the embodied, encultured, reciprocal, dynamic, living process of human learning that takes in a myriad of different sources.
In a legal sense, you are a person consciously composing each element of the song. If you emulate a pre existing song, you most likely do so consciusly. If you want to borrow a sample, you clear it. If you do a cover, you credit the original songwriter. And when you perform it, it expresses your personality: your limitations and idiosyncrasies as a performer, your choice of instrument, acoustics, effects. Your subjective interpretation of the emotional resonance and meaning of the song.
Whereas model is nothing *but* samples. A statistic compilation of underlying works, from which you can extract derivatives.
I don’t really see any similarity at all. Not ontologically, not technically, not legally, not morally.
…see more
The whole discussion is here, and it goes pretty deep: https://www.linkedin.com/in/stevestewart7/recent-activity/comments/. You may have to scroll down a bit to find it...
There. Intro edited to acknowledge your contribution. Let me know if you have notes. Thanks for being a good sport!
Thanks, Steve! I did paraphrase your parts, and I did tell you about this one. Wasn’t sure you would appreciate the attribution, but now that I know you want it, I’ll of course include it. Cheers!
An alternative to Adobe is attracting some attention - Affinity by Serif (now owned by Canva). Currently 50% off for a perpetual multi-platform, 3-tool bundle: https://affinity.serif.com/en-us/affinity-pricing/
I learned about it from Evan Winston here: https://www.linkedin.com/posts/evan-winston-750090b7_affinity-adobe-noadobe-activity-7205250349945790465-Hn0g
It’s really good! I don’t trust Canva though
Hi Johan :) I read through Affinity’s T&Cs and didn’t see any red flags. I haven’t checked Canva’s yet (don’t use it). New corporate parents can definitely drive changes. What makes you most uneasy about Canva?
Ok, so I had a hunch I had heard something bad, and sure enough: Canva provide Stable Diffusion image generation in their products. Stable Diffusion is a research model not licensed for commercial use. They also partnered with OpenAI. So even if their products don’t seem that data greedy today, the parent company is rotten.
Ah … good to know. Thanks for sharing that info.
No particular reason tbh. Just haven’t done my research.