🐝
Markov

@Markov

$0total balance
$0charity balance
$0cash balance

$0 in pending offers

Comments

🐝

Markov

about 2 months ago

@Ryan Kidd Hey Ryan! The major updates are -

  • We have built a new website! Which incorporates a bunch of functionality! There are a many of details around this, but the core things are:

    • Smooth workflow & Fast Iteration: We wanted to be able to both collaborate and get granular line by line feedback on our writing, but still provide a clean experience for students who simply want to read the text. So we structured the site/workflow to work, collaborate and comment on Google Docs, and then a simple export makes the content appear on the site. This system makes versioning, updates, and overall life easy for both commenters and us as writers. It allows for super easy incorporation of new papers and corrections over time, making it more of a ‘live’ project.

    • Interactivity: We have embedded content from places like Metaculus, that gives students live updates from prediction markets for things like timelines or takeoff, for example. Instead of just being static images.

    • Collaborations: Besides this, we are officially collaborating with Stampy (AI Safety Info) where we provide detailed technical content, and they provide smaller summaries! So, as just one example of added functionality due to this collaboration, the website has on-hover definitions for all jargon sourced from Stampy. Similarly, we also spoke to collaborate with other projects like AI Digest, to incorporate more interactive snippets from their website that adds value to the site in addition to the core text.

    • Plug and play: We now provide both online reading, offline downloads, and have facilitation guides for most chapters. This gives any AI Safety course and reading group the ability to just pick up and run with the project extremely easily! (We are looking to scale this type functionality even more)

    • More modalities: We are experimenting with both audio and video (due to popular requests from students) using things like NotebookLM, and also have plans to record talks/presentations as video alternatives to reading the chapter.

  • Core Text: As for the core text, we have written a lot! We have updated some older chapters, and written new ones. We will have a v0 of every chapter on site soon. Currently, we have just 3 chapters left. They are all written but still need to get refined before appearing on site. We will be pushing them very soon!

  • Increasing students: We have new courses reaching out to use the content often.  We have steadily increasing numbers of students, and the projections for how many more we expect are really exciting! Some current courses, and number of students per year:

    • ML4Good (250)

    • AI Safety Collab (80)

    • ENS Ulm + Saclay (100)

    • UBC AI Safety, in Vancouver (~30-50)

    • AI Safety Goethenburg

    • AI Safety Hungary

    • Independent readers

  • Impact metrics and analytics: Besides the courses already using the text, we want to be a little bit more formal in getting impact metrics. There are feedback forms that we get from some courses, but we plan to incorporate analytics into the website, and a feedback form directly at the end of every chapter. This lets us see how many unique students we are getting, and lets people who might not be directly affiliated with a course, but are using our text, give feedback.

  • Based on talks at various conferences (EAG/EAGx/...) both experts and students think that this is a great project, and have found value in even the current version of it built out so far.

  • We are in talks with senior architects to get more people formally on board.