Impact track record

The best predictor for the work that will happen at Condor Camp South Africa is the work we've accomplished with our team in Brazil. Over the past two years, we've provided a significant portion of the infrastructure to support talented students moving from AIS-curious to deliberately pursuing it as a career path in Brazil. We are asking for funding to expand our reach to other continents in the Global South, starting with South Africa, in the form of a 7-day immersive AI Safety career accelerator camp.

In brief, during that time we:

• Ran three successful Condor Camps for Brazilian university students, an ML4G for Latin American students, amongst other short programs, with great results in terms of career transitions and engagement in AI Safety

• Developed an effective curriculum for introducing AI Safety concepts to talented students in the Global South

• Created a network of alumni who are now actively contributing to the AI Safety field

Key outcomes from previous Condor Camps, in total 47 participants:

• 4% are in AI Safety positions (e.g., GovAI research assistant, SPAR fellow)

• 23% engaged in AI Safety research (e.g. replicability, Optimal Control Theory applied to AI safety) 

• 25% working on AI Safety community building

• 47% found the Camp very helpful for their careers

• 32% shifted towards AI Safety careers

• 82% of participants rated the likelihood of achieving similar outcomes without the Camp as "very unlikely" or “unlikely”. Suggesting Condor Camp's uniquely significant impact on participants' career plans that wouldn’t have happened otherwise.  Note that this question was only asked at the latest camp.

• 2 students engaged with research created a Handbook in Mechanistic Interpretability, and managed to start a class on MechInterp on their highly prestigious Bachelor in AI institution (FGV Rio).
  

We conducted a 6-month follow-up for the latest cohort (CC3) and a 12-month follow-up for the cohort before (CC2), the response rate was 70%. 

Here are some relevant results:

• 76% agreed or strongly agreed that the camp contributed to their progress in AI Safety (90% in the latest cohort!)

• 87.5% rated 4 or above (out of 5) the usefulness of the camp activities and opportunities, in hindsight

Plan for impact

Our basic plan for impact is to reduce risk from artificial intelligence by building a pipeline of talented individuals from the Global South who are equipped to work on AI Safety. Here's our current best concrete plan:

Our simplified Theory of Change

1. Find talented people
   Find talented students curious about AIS but don’t have a clear path towards a meaningful career in the space.  

2. Inform and equip
   Bring talented students up to speed on the AIS landscape through informative sessions with subject-matter experts and provide career coaching & introduction to career paths to equip them with the tools they need to take concrete steps towards a career in AIS. 

3. Help students connect the dots
   Partner with AIS Cape Town to continue providing students with support post-camp, encouraging them to continue building relevant skills, connecting them with opportunities, and providing social motivation & accountability so they’re applying for relevant opportunities within 6-12 months. 

Of the projects Condor Initiative works on, you are probably most familiar with our Camps in Brazil. The funding we're asking for is both to bring this successful model to South Africa and to work on more infrastructure that's as valuable to the AI Safety ecosystem in the Global South.

Expanding to South Africa… and the Global South?

Expanding to South Africa represents a strategic opportunity to diversify and strengthen the global AI Safety talent pipeline. We believe that the talent pool in Africa is significantly underserved in terms of AI Safety outreach and education.

By bringing Condor Camp to South Africa, we aim to:

1. Tap into a new pool of highly talented individuals

2. Create local momentum for AI Safety work in Africa

3. Diversify the perspectives and approaches in the global AI Safety community

4. Build connections between the AI Safety communities in South America and Africa

Ultimately, we are planning to build whatever infrastructure will most help people who are successfully reducing existential risk from AI in the Global South, contributing to a flourishing future for humanity.

Activities in need of funding:

Our budget for Condor Camp South Africa is approximately $70,000, of which we're seeking $25,000 in additional funding. The Condor Initiative has committed USD 50,000 to this project. We want to ensure the long-term sustainability and potential expansion of the program.

Rough breakdown

22 students, 4 international speakers, 2 local speakers, 3 volunteers, 4 organisers.

7 full days of content, 8 nights.

$47,000 | Accommodation, meals, travel/transportation

• Venue stays: $17,000

• Conference rooms, meals, and coffee breaks: $15,000

• Flight tickets & land transport $14,000

• Liability insurance: $1,500

$14,700 | Speakers, fellows, and operational costs

• Speaker honorarium: $7,000

• SA Organizing team: $5,500

• Goodies and support materials: $2,200

$8,000 | Fellows support (post-camp project grants)

• Fellows: $8,000

$6,000 | Contingency buffer (not priced in)

• 10% of total budget: $6,000

How Condor Camp works

One of the most useful facts to understand about Condor Camp is our commitment to adapting our approach based on local context and participant feedback. We iterate on our curriculum and program structure after each Camp, ensuring that we're providing the most relevant and impactful experience possible.

The program structure counts career workshops, lectures (called masterclasses) on technical alignment and governance, peer-to-peer knowledge sharing through lightning talks, hands-on exercises with speakers' mentorship, project development and management, amongst other activities, with a specific focus on the linearity we've learnt from other Camps.

Our team consists of generalists who can take on various roles as needed, allowing us to pivot and adjust our strategy as we learn more about the specific needs and challenges in the South African context.

We also prioritise building long-lasting relationships with our alumni, creating a network of AI Safety-minded individuals across the Global South. This network effect amplifies the impact of each individual Camp.

What are the most likely causes and outcomes if this project fails?

Below are some of the most important ones, and we're happy to go into more detail about the failure modes upon request.

No conversion, ideas too weird

Mitigation: Focus on scientific presentation, involve respected organisations, encourage criticism and red teaming.

Not tapping into top talent

Mitigation: Outreach to top universities (also the top universities in Africa as a whole), consulted with AIS talent recruiters from similar programmes, design a rigorous selection process.

Spark interest but not engagement

Mitigation: Provide post-camp fellow grants, involve AIS Cape Town to absorb students after the camp, foster group cohesion, and maintain active post-camp communication.

Camp doesn't translate to the South African context

Mitigation: Involve local South African advisors and organisers, and iterate on the selection process based on feedback.

No counterfactual impact

Mitigation: Design the advertisement and selection process to select students who are new to the field of AIS.