BASE Technical AI Safety Training Initiative

Raising $50,000active

Providing GPU credits and instructional support for 40 participants completing the ARENA AI Safety curriculum through Black in AI Safety and Ethics (BASE)

Endorsed by

Ryan+1

Donate:Manifund

Providing GPU credits and instructional support for 40 participants completing the ARENA AI Safety curriculum through Black in AI Safety and Ethics (BASE)

Endorsed by

Ryan+1

Donate:Manifund

Project Details

Black in AI Safety and Ethics (BASE) will provide technical training in AI safety to 40 participants from the Black diaspora. Twenty participants will complete a 12-week remote ARENA training program beginning July 19, and twenty participants will complete a 5-week remote technical training program during Cohort 2 of the BASE Fellowship beginning September 8. Eleven participants completed the 12-week ARENA training ahead of Cohort 1, and 17 completed the 5-week technical training during the cohort.

Interpretability, evaluations, oversight, and other technical AI safety exercises cannot reasonably be completed on a personal computer. Providing GPU credits will allow participants to work through those exercises hands-on throughout both training programs, rather than learning the material solely through theory.

The project will be led by Lawrence Wagner and Robert Amanfu, with two teaching assistants providing office hours, debugging support, assignment support, and technical guidance throughout both training programs. The concrete output will be 40 participants who finish the technical AI safety training, including the compute-intensive exercises that are part of both programs.

Theory of Impact

Advanced AI systems will require more researchers with the technical skills to evaluate, interpret, and identify potential risks before deployment. One of the biggest barriers to developing that talent is access to compute. Our project develops that talent within the Black AI safety community, where ability and motivation already exist but access to compute does not. By providing GPU credits, participants can complete the technical AI safety exercises necessary to build those skills. More researchers with hands-on experience in AI safety give the field more people who can help evaluate advanced AI systems and reduce catastrophic risks from AI

People

Lawrence Wagner

creator

Grants Received– no grants recorded

Add grant

Funding Asks

grantmaking.ai Launch Round

Applied

Minimum

$50,000

Ideal

$150,000

What this ask is

How the money will be spent

Funding will be used for:

GPU credits for 40 participants: $27,000
Teaching Assistant 1: $7,500
Teaching Assistant 2: $7,500
Program coordination and technical support: $8,000
Total request: $50,000

GPU credits will allow participants to complete compute-intensive exercises throughout the ARENA curriculum, including interpretability, evaluations, oversight, and other technical AI safety work that cannot reasonably be completed on personal computers

Discussion

Sheikh Abdur Raheem AliEndorsement14d

Nurturing the next generation of talent is a no-brainer. I do want to flag that ARENA doesn't really require that much GPU compute, and the exercises which do aren't core to the curriculum. I'm neutral but would endorse if you change to paying the participants an unrestricted stipend/gift and let them decide how they want to use it unless that gets complicated for legal reasons.. This would also help attract higher quality applicants with optionality over opportunities.

Lawrence Wagner13d

Thanks for flagging this. I checked both the ARENA website and our technical trainer to better understand the course’s compute requirements. About 30% of the notebooks include GPU-dependent work, around 20% are effectively GPU-bound, and some newer alignment exercises do not run reliably on free Colab. Reliable compute matters most during the hands-on portions of the course, where participants complete capstone projects, collect activations, work with SAEs, fine-tune models, and experiment with larger open models.

Based on your feedback, we revised the budget to reduce the compute allocation and direct more funding toward participant stipends:

GPU credits for 40 participants: $11,000
Participant stipends: $16,000
Teaching Assistant 1: $7,500
Teaching Assistant 2: $7,500
Program coordination and technical support: $8,000
Total request: $50,000