Why we work in public at BlueDot Impact – BlueDot Impact

Why we work in public at BlueDot Impact

By Adam Jones (Published on June 16, 2024)

The availability of great public resources has wide-ranging benefits: both for the people publishing the content, and for others who want to use it. We think people often underestimate these benefits, but minor inconveniences like hiding information behind an email request turns a lot of people away. This is likely even more true for people new to the field (like many of our course participants!), as they are less likely to feel part of your community and therefore able to ask you for help.

At BlueDot we’re going to trial publishing more of our internal[1] resources, systems and datasets.[2] We hope that this will:

  • Make it easier for people to find relevant information. This will help us too, but we expect this will be particularly helpful for potential applicants, course participants and partner organisations to understand how we operate.
  • Free up our time to focus on important work. Given people can find answers themselves, we’ll receive far fewer requests for information - plus we can redirect more questions to things we’ve published.
  • Help develop the AI safety and biosecurity fields indirectly, for example because people can use our tools to run local versions of our courses, or learn from our experiences.
  • Further develop these fields more directly, for example by publishing gaps in literature we’ve identified.
  • Promote AI Safety Fundamentals and Biosecurity Fundamentals as insightful authorities in their respective domain areas (while appropriately communicating uncertainties we have in these fields).
  • Promote BlueDot as a great place to do high-quality work on impactful problems.

Over the last few weeks, this has included:

  • Publishing the session plans for all of our current flagship courses, and all the courses we ran in 2023.
  • Publishing course-specific facilitator guidance for all of our current courses.
  • Publishing our facilitator training course, including the curriculum, exercises and session plans.
  • Releasing all of our custom software code publicly, licensing it under a license approved by the Open Source Initiative, and making issues and feature requests public. Longer term we’ll also look at making all the documentation public, as well as improving the systems so they’re both: (1) easier for others to run themselves, (2) some are provided as a hosted SaaS offering.
  • Publishing key internal documents we reference a lot internally, including our Airtable standards.

As we do this, we’ll still be only publishing things we think could be genuinely useful or valuable to others - not just AI-generated mush. We’ll also be making sure that we’re clearly communicating our confidence in our beliefs, and titling posts accurately so it’s easy to quickly identify whether content is relevant.

We’re keen for people to contact us with feedback if they’ve engaged with things we’ve been publishing under this initiative (whether positive or negative!).

Footnotes

  1. As well as publishing our internal stuff, we’ll be encouraging others to publish more. For example, on our recent AI alignment project sprint, we’ve encouraged people to work more publicly. This includes things like publishing partial progress reports and documenting some of the more mundane difficulties of running a project and how people got around them.

  2. Starting to follow in the footsteps of other organisations who have done this, including GitLab and getonboard.

We use analytics cookies to improve our website and measure ad performance. Cookie Policy.