Good Futures Initiative

Use your winter break to make a difference!

Begins December 18th and ends by January 25th. Applications close December 11th.

Time commitment: Minimum 12 hours per week, remote.

Good Futures Interns will begin a research or other high-impact project over the course of winter break. Projects will accomplish one of these three goals while producing an end product (e.g.: a blog post sharing your research findings) :

  1. Skill up the intern in order for them to work on AI Safety or Biosecurity in the future.

  2. Let the intern explore an aptitude for an impactful career.

  3. Create impact directly.

In addition to their focused project, students will develop a more general understanding of project management through weekly goal setting and check-ins, attending talks/Q&As with project managers and researchers, and spending 1 hour per week skilling up through readings while completing their project.

Our focus is on providing an environment to set up students and recent graduates to pursue impactful work in the future. This means we'll be making application decisions based on how much the project creates impact in the present and sets the applicant up to do impactful work in the future.

Types of Projects

Don’t feel qualified but still want to do something impactful? Still apply! When you apply to take on a project, you can click a box that will let us know if you’re interested in being put in a cohort for accountability even if we aren’t able to give you funding and the whole experience.

Propose your own research or project idea with the project proposal form. First, email Aris (alarichardson@berkeley.edu) before applying. If you have skills you want to develop or an aptitude you want to test out but aren’t certain how, Aris can collaborate to plan a project for you to work on!

Pick a project from the example projects below to apply to work on. Apply with our project proposal form by writing details about how you would execute the project.

Support: $300 stipend and up to $1,000 of funding for your project.

Community: Join a cohort of ~6 students working on other independent projects and have weekly meetings to share progress with your cohort.

Expertise: Listen to advice from researchers and other professionals working on reducing global suffering and preventing existential risk.

Accountability: Weekly one-on-one check-ins to keep you on track, increase the impact of your project, and prevent failure modes.

We’re most interested in supporting projects on AI Safety and Biosecurity.

We are also accepting applications about the following topics, although they will need to be competitively impactful to be accepted:

  • Reduction of existential risk through nuclear power or extreme climate change

  • Increasing people’s ability to emotionally connect to longtermism

  • Farmed or wild animal welfare 

  • Effective policy and advocacy  

  • Effective giving

Example Projects

You can choose to propose projects listed below to apply to take on, or propose your own project of the same caliber of effectiveness. Our project proposal form for the application can be found here or in the application itself.

Skilling up: Technical AI Safety Project Examples

Review GPT-3 responses and create a blog post about your findings

Consider questions where the most common answer online is likely to be false (as in TruthfulQA). If you prompt GPT-3-Instruct (or similar) with questions and correct answers in one domain/topic, then ask it about another domain/topic, will it tend to give the correct answer or the popular answer? As you make the domains/topics more or less different, how does this vary?

Mentorship: Sam Bowman is open to receiving questions via email: bowman@nyu.edu (no commitment to respond) 

Research Mechanistic Interpretability and create a blog post about your findings

Project 1: Identify an interesting neuron in a one-layer language model at neuroscope.io and investigate exactly what inputs do and do not make that neuron activate (demo notebook)
Challenge: Reverse engineer how the model computes that neuron with the TransformerLens library (Tutorial)

Option 2: Project 2: The Interpretability in the Wild paper (https://arxiv.org/abs/2211.00593) did a detailed reverse engineering of the circuit GPT-2 Small uses to figure out that the sentence "John and Mary went to the store, then John handed a bottle of milk to" ends with " Mary" rather than " John". Inspired by this, find an interesting, concrete task that a language model can do, and then identify which parts of the model are used to compute this. (demo notebook to crib)
Challenge: Use the TransformerLens library to reverse engineer how the task is computed (Tutorial). See also the paper's codebase.

Mentoring Message: Feel free to email me (neelnanda27@gmail.com) with what you find (no commitment that I'll have time to respond, sorry!)

Exploring an Aptitude / Creating Impact Examples

Researching social factors that contribute to existential risk - Academia aptitude

An example of such work is this report examining the effects of prestige during the space race and its implications for global coordination when transformative AI is developed.

This project would be a good fit for someone who has experience with historical, anthropological, or psychological research.

Start an Existential Risk Initiative at your school - Community Building Aptitude

This project would be good for someone who is interested in AI Safety, Biosecurity, or community building. This could be a way to test out community building.

Translate useful texts about Effective Altruism into other languages.

For example, translating 3 relevant research papers from AGISF into Mandarin and posting them somewhere where they can be accessed by ML engineers in China.

This project could be a good fit for someone who is studying languages or international cultures or comes from a non-English background.

You can also get inspiration for your own projects from Apart Research Hackathon winners.

Creating visualizations of ideas about longtermism and the future of human life.

This project would allow people to more easily connect with concepts about the long term future and human lives in factors of billions.

This project would be a good fit for someone who has graphic design or data visualization skills.

Distill research and publish your work - Communications Aptitude

For example, create 2 articles distilling longtermist ideas and submit them to newspapers/magazines. 

This would be a good fit for someone with a research interest in English, writing, or communications.