So—what have we been up?
We released 46 hours of human narrations:
- 71 narrations of the core readings from the AI Safety Fundamentals courses.
- 14 narrations for the new 80,000 Hours Career Guide and their infosecurity career review.
- 10 curated posts for EA Forum
- 11 curated posts for LessWrong
That's an average of roughly 30 minutes per day, every day during Q2.
Industry standard is 2-5 hours’ work per hour of finished narration, depending on the difficulty of the material. So yeah—Perrin has been smashing it, with the help of Katy and Marcus on proof-listening and quality assurance.
We estimate these narrations will reach >100K listeners by the end of the year.
Our AI narration system is finally ready for prime-time!
We ran a pilot with the Effective Altruism Forum, narrating all new posts during June and releasing them on the podcast feeds we manage on their behalf.
We narrated some >200 posts in June, and these received roughly 500 hours of listening during the month. The EA Forum team were pleased with the results, so we'll continue the partnership for the rest of 2023. We’ll continue narrating all new posts, and we’ll also narrate the best posts from their archive.
In July, we’ll start a pilot of the same service for LessWrong.
We also narrated the “Introduction to Utilitarianism” textbook. This is the first test of our “human edits, AI narrates” service, where we manually edit and proof-listen to get the very best results from the AI voice models.
We released version 1 of our audio player. Five of our clients are using it on their websites. It has been loaded >500K times, and served >1500 listening hours.
It is already the best audio player on the internet for many use cases. We're working on version 2, which will make it #1 on all dimensions for most use cases. 😎
TYPE III LIBRARY
We released a scrappy alpha version of our website, which will help people find great things to listen to.
It's extremely nascent right now—it wasn't a major focus in Q2. The embarrassing alpha will become a respectable beta during Q3. 😅
We decided that the “You record, we everything” service should not be part of TYPE III AUDIO’s offering. Jason Cotrebil left TYPE III AUDIO to develop this service as a separate company.
In April we completed a 1 week work trial with a very promising candidate. In July, this person began a 3 week project to develop the TYPE III LIBRARY.
Shortly after starting our fundraising round, I realised that it wasn't actually the right time to seek investment. The reason is: I'm still exploring other project ideas for 2024. The first year of TYPE III has always been about assessing the upside case for this project, with the plan to review in autumn and decide whether to double-down. I want to make that call before raising significant external investment. So, for now, we bootstrap.
Alongside my work on TYPE III AUDIO, I spent about 3 weeks full-time equivalent learning about AI. I read some 100 papers and blog posts, spent a couple weeks in Berkeley, CA, and did dozens of 1-1 conversations and Google Doc exchanges. My understanding of the “state of AI” is an important input to the success of TYPE III AUDIO, and also to my thoughts on what to work on in 2024.
I'm pleased with the progress we made in Q2.
Our team size during this period was roughly 3 full-time-equivalent.
Plan for Q3
Create hundreds of hours of AI narrations every month. Release the beta of TYPE III LIBRARY. Continue releasing human narrations.
You get the idea.