Alignment Research Fellowship (ARF)

As AI systems become increasingly capable, there's a growing need for technical talent in AI safety research. Many aspiring researchers find themselves asking

"What's the next step after learning about AI safety fundamentals?"

This program provides structured training in technical AI safety through hands-on implementation of key machine learning concepts and alignment research techniques and is based on the ARENA (Alignment Research Engineer Accelerator) curriculum.

Program Details:

We are excited to announce our online technical upskilling pilot program in collaboration with the European Network for AI Safety (ENAIS), and AI Safety Hungary. You'll master key technical concepts in AI safety through structured co-working sessions and hands-on implementation, with most of your time spent under the guidance of experienced Teaching Assistants (TAs).

The program runs online starting on the week of the 22nd of March and includes:

• Two 3-hour co-working sessions per week with guidance from experienced TAs

• 4 hours of structured independent study per week

• Small group sessions with a dedicated TA

• Ongoing support via Slack

• Total commitment: 10 hours per week for 14 weeks from March to June 2025

The program is completely free to attend. We provide:

• Teaching Assistants

• Compute credits for projects

• Small group collaboration opportunities

Curriculum Structure

• Based on the ARENA materials, you'll work through:

• Building a transformer from scratch and understanding attention mechanisms

• Deep learning fundamentals and PyTorch implementation

• Mechanistic interpretability of transformer models

• Introduction to Reinforcement Learning and RLHF

• LLM evaluations: designing benchmarks and building evaluation frameworks to assess model capabilities