I have set up a philosophy podcast! It was far easier than I thought it would be. Each episode is a chat about an article from the SEP — The Stanford Encyclopedia of Philosophy. Hence the title, ChatSEP. I believe that this will be the primary podcast that I listen to for the next year. It really makes concrete the possible future where everyone makes their own bespoke entertainment.
Here is a link: https://open.spotify.com/show/2pUm5OK4sG8RzDDcXUhOvQ?si=979e18a49090472d
And to the SEP article: https://plato.stanford.edu/entries/descartes/
I am releasing 7 ten-minute episodes per day and will thereby cover all of the 1803 SEP articles in about 9 months. So far the episodes are about philosophy of religion, medieval philosophy, and most recently a series on Descartes. Later episodes will be about philosophy of science, logic, aethetics, ethics, Kant, Plato… literally every topic in philosophy. Here is a random SEP article.
I have already generated all of the content (350 hours, 60Gb) using Google’s NotebookLM. After automating the procedure it took only 2 minutes to generate each episode, so 60 hours total. It then also takes about 2 minutes to schedule each episode to appear on Spotify. (I am doing this manually atm). Of course, I shouldn't take any intellectual or artistic credit for these things myself. The human effort that went into this project is as follows:
- scraping links to all the SEP articles,
- writing up a paste-in source to give it authorship information (this doesn't appear on the SEP article itself).
- scraping which article reference each other,
- coding up a nice walk through this graph (i.e., the episode order),
- automating the content generation,
- manually scheduling the episodes.
I find two things crazy. Firstly, the quality of these episodes. Secondly, how one man in the course of a week can generate a years worth of podcast content. As I noted above, I believe that this will be the primary podcast that I listen to for the next year. It really makes concrete the possible future where everyone makes their own bespoke entertainment. And right now is the worst that this tech will ever be. Crazy!
In the future, will everyone make their own bespoke entertainment? A Case Study
byu/Main_Scratch6399 inFuturology
4 Comments
Submission Statement: As I noted in the post, this project makes concrete the possible future where everyone makes their own bespoke entertainment. In fact, I believe that this will be the primary podcast that I listen to for the next year. I find two things crazy. Firstly, the quality of these episodes. Secondly, how one man in the course of a week can generate a years worth of podcast content. And right now is the worst that this tech will ever be. Crazy!
we can do that already. People like to get stories from other people, that is sort of the point. Then they share that experience with others. We already have personalised self made stories, it comes free with our mind.
You still have to listen to all the podcast. Let us know at the end how repetitive and boring it came out, because I think it’s sincerely an expected outcome
First impressions: this is pretty neat!
As someone with a background in both philosophy and computing this hits a sweet spot for me. Doubly so since I love to learn and have a lot of time for just passively absorbing audio content (I hike/walk a lot).
The issues come when I start thinking about this more deeply (or frankly, even at a practical level).
1. The SEP articles are right there, and while some can be a little dense, they aren’t terribly long. IEP is also pretty decent and sometimes can be a little more digestible. There could be an argument made that blending the two gives you richer content and something above and beyond what they individually provide.
2. Okay, so you could just read them and a summary isn’t that helpful, but it’s a podcast! Sure, having them read to you is great, but text-to-speech is often a built-in and doesn’t really require AI or additional tooling, so what’s the added advantage of podcastifying these articles?
3. This is my main takeaway: the thing that makes or breaks a podcast isn’t so much the information (which your solution clearly provides) but the presentation. That will be tricky to get right with AI at the moment, but maybe future iterations will have more personality. I can imagine this working well once or twice, but if I’m going to have articles read to me, I’m going to need something more engaging than text-to-speech, and that’s something even humans can’t always get right.
It’s an interesting idea, but I wonder to what extent the technology can support it. This is a not question about whether AI *can make art* or even whether what it makes *counts as art*, but whether what it makes is *worth consuming*. The issue is that we haven’t really figured out how to consistently reproduce engaging media, and it’s not clear that this is something that can be straightforwardly gleaned from massive training sets. Until we crack the “what makes media good”, the best we can hope for is something low to mid quality that’s easy, which worries me in general, because it hints at a future where we consistently trade off quality and meaning for speed and convenience. That’s a fine trade off for a lot of industries (fewer than you’d think when you get right down to it), but it seems entirely antithetical to art and entertainment. Now whether this is purely educational and therefore outside the bounds of these concerns raise other concerns about what constitutes good teaching.
Anyway, good effort!