all/meta, though I think the main value add is in AI
(Nonlinear has made a great rough/low quality version of this, so at least some credit/prize should go to them.)
Audio has several advantages over text when it comes to consuming long-form content, with one significant example being that people can consume it while doing some other task (commuting, chores, exercising) meaning the time cost of consumption is almost 0. If we think that broad, sustained engagement with key ideas is important, making the cost of engagement much lower is a clear win. Quoting Holden’s recent post:
I think a highly talented, dedicated generalist could become one of the world’s 25 most broadly knowledgeable people on the subject (in the sense of understanding a number of different agendas and arguments that are out there, rather than focusing on one particular line of research), from a standing start (no background in AI, AI alignment or computer science), within a year
What does high quality mean here, and what content might get covered?
High quality means read by humans (I’m imagining paying maths/compsci students who’ll be able to handle mathematical notation), with good descriptions of diagrams. If posts involve conversations (e.g. the MIRI logs), different voices are used for different people. Holden’s cold takes read throughs are a good example.
High quality also means paying for or otherwise dealing with copyright, curating pieces into much more searchable/navigable collections than the current podcast feeds.
What sort of things?
alignment forum posts, with sequences collated into playlists.
The MIRI conversations
Key technical reports e.g. Carlsmith on power seeking AI, Cotra on Bioanchors
New books (The Long View)
everything on key reading lists, again organised into playable feeds. e.g. Tessa’s biosec list, jtm’s longtermism list, the AGI safety and governance fundamentals curricula, the introductory and in-depth fellowship reading materials.
Frankly, I’d like the ability to send a written feed to somewhere and have it turned into audio, maybe crowdfunded. Clearly non-linear can do it, so why can’t I have it for, say, Bryan Caplan’s writing.
High quality, EA Audio Library (HEAAL)
all/meta, though I think the main value add is in AI
(Nonlinear has made a great rough/low quality version of this, so at least some credit/prize should go to them.)
Audio has several advantages over text when it comes to consuming long-form content, with one significant example being that people can consume it while doing some other task (commuting, chores, exercising) meaning the time cost of consumption is almost 0. If we think that broad, sustained engagement with key ideas is important, making the cost of engagement much lower is a clear win. Quoting Holden’s recent post:
What does high quality mean here, and what content might get covered?
High quality means read by humans (I’m imagining paying maths/compsci students who’ll be able to handle mathematical notation), with good descriptions of diagrams. If posts involve conversations (e.g. the MIRI logs), different voices are used for different people. Holden’s cold takes read throughs are a good example.
High quality also means paying for or otherwise dealing with copyright, curating pieces into much more searchable/navigable collections than the current podcast feeds.
What sort of things?
alignment forum posts, with sequences collated into playlists.
The MIRI conversations
Key technical reports e.g. Carlsmith on power seeking AI, Cotra on Bioanchors
New books (The Long View)
everything on key reading lists, again organised into playable feeds. e.g. Tessa’s biosec list, jtm’s longtermism list, the AGI safety and governance fundamentals curricula, the introductory and in-depth fellowship reading materials.
Frankly, I’d like the ability to send a written feed to somewhere and have it turned into audio, maybe crowdfunded. Clearly non-linear can do it, so why can’t I have it for, say, Bryan Caplan’s writing.
If you’re ok with autogenerated content of roughly the quality of nonlinear, both Pocket and Evie are reasonable choices.
High-quality human performance is much more engaging than autogenerated audio, fwiw.
Hence the original pitch!
Non-Linear could be paid to repost the most upvoted posts but with voice actors.