In June we started offering AI narrations for all new EA Forum posts.
The feature proved popular[1], so we’ve now enabled narrations for all >100 karma posts in our archive. You can listen via the post pages on our website.
In addition: every Monday and Thursday, our podcast feeds will feature a “classic post” from the archive.
We will continue making narrations for all new EA Forum posts when they are published. And, we will stop making human narrations for curated posts[2].
How to listen
You can listen to post narrations on our podcast feeds, and on post pages.
On our podcast feeds:
 | EA Forum (Curated & Popular) Audio narrations from the Effective Altruism Forum, including curated posts and posts with 125+ karma. Subscribe: Apple Podcasts | Spotify | RSS | Google Podcasts (soon) |
 | EA Forum (All audio) Audio narrations from the Effective Altruism Forum, including curated posts, posts with 30+ karma. Subscribe: Apple Podcasts | Spotify | RSS | Google Podcasts (soon) |
On post pages:
Some improvements
We’ve added chapter markers to the episode descriptions, so you can see the post outline and skip to sections of interest.
Sometime soon we’ll probably try releasing a small number of other kinds of episodes on the podcast feeds, e.g. a single episode that contains all posts from a sequence.
And, we’re working to improve the quality of the narrations.
How could we make this better?
Thank you to everyone who already shared feedback on this feature.
How could we make this better for you? Let us know in the comments, by email, or by clicking the “Feedback” button on the audio player.
Appendix 1. What else would you like to listen to?
TYPE III AUDIO will release hundreds of hours of AI narrations in the next 1-2 months. We’re currently thinking about what to prioritise.
What would you like to listen to? Post below or send us an email.
In the meantime, you can find a bunch of interesting audio on the alpha version of our website.
Appendix 2. Want to listen to Google Docs and academic PDFs?
The latest version of Speechify is pretty good.
- ^
Total listening time was roughly 400 hours in June and 300 hours in July. The podcast feeds have several hundred subscribers. Qualitative feedback has been very positive overall.
- ^
Human narrations are expensive ($500 per finished hour of audio). Our data suggests that AI narrations receive a similar completion rate. The main weakness is that some posts really benefit from a human-written summary of graphs, tables and images. We may experiment with “human-edited AI narrations” to mitigate this issue for curated posts.
You don't have copyright permission for old posts, right?
Hi Kirsten, thanks for raising this issue. You're right, we don't have copyright permission. I'm going to be emailing authors of the specific set of posts we plan to narrate from before the license change.
Thanks for clarifying!
I'm curious why people downvoted this comment! (when I posted this, it was at 0 with four votes, and I strongly upvoted it to 7). I think it is an important question and is currently unanswered. For reference on its importance — it's directly relevant to me in a context related to doing my work for an EA organization, and in particular trying to catalogue historic IP.
Thanks for adding this feature!
I am also interested in how this is structured from a licensing perspective - this is relevant for content posted with permission, but not owned by the original poster (which is relevant in some cases I'm looking into), and also for people's old content generally. Would the Forum team be able to clarify who owns the audio versions, and what the licensing on pieces posted prior to the current terms of use was? My impression was that they were owned by the authors, but I can't find any records of the terms of use prior to this version: https://forum.effectivealtruism.org/termsOfUse. But, I also have old content I posted that now has a recording, and didn't get asked about this use of it, so am assuming I actually had granted permission to EV to use it at some point? (and I am happy for it to be used this way). If you happen to have a copy of the terms of use from prior periods, I'd be interested in seeing them.
Also, this is minor and somewhat unrelated, but I noticed it while looking into this — in Ben's announcement post, it says that the CC-BY license kicks in on December 01, 2022, but the current terms of use suggests that it is in place as of October 31, 2022. I'd also be interested in knowing what license governs pieces posted between October 31 and December 1 right now.
(BTW, I'm excited about having the audio recordings, but just trying to work out how the licensing on this works).