How to apply for Alignment Scientist / Engineer

About AE Studio

AE Studio is a boutique software development and data science firm that prioritizes high-impact projects and a culture of continuous learning. Their focus on AI alignment reflects a commitment to building safe and beneficial AI systems, making it an ideal place for researchers who want their work to have real-world ethical impact.

About the role

As an Alignment Scientist/Engineer, you'll conduct independent research on AI safety, focusing on pretraining, interpretability, and model safety. Your experiments will directly contribute to the development of safer AI systems, and you'll collaborate with external researchers to shape the direction of alignment research.

A typical day

A typical day might involve reading recent alignment papers in the morning, then running experiments in PyTorch to test a hypothesis about model representations. You'd collaborate with a team member over Slack to analyze results, and later draft a section for a research paper or prepare a presentation for an external collaborator.

Who AE Studio is looking for

• Has a strong publication record in AI alignment or related fields (e.g., mechanistic interpretability, adversarial robustness).
• Is highly proficient in Python and PyTorch, with experience building and training deep learning models from scratch.
• Thrives in a research environment where they decompose complex problems into testable hypotheses and iterate quickly.
• Can communicate technical concepts clearly to both technical and non-technical stakeholders, and is eager to contribute to research publications.

Tips for this application

• Tailor your resume to highlight specific alignment research projects, such as work on activation steering or sparse autoencoders, not just general ML experience.
• Include a link to your GitHub or personal website with code from alignment experiments, preferably in PyTorch.
• In your cover letter, mention a specific paper or approach in alignment that excites you and explain how it relates to AE Studio's work.
• If you have any experience with LLM safety or RLHF, explicitly call it out, as it's highly relevant to pretraining and model safety.
• Prepare a short (1-2 paragraph) research statement outlining a question you'd like to explore in the role, showing independent thinking.

What to cover in your cover letter

['Emphasize your hands-on experience with deep learning experiments and your ability to design testable hypotheses.', 'Show genuine passion for AI alignment and mention specific alignment challenges you want to solve.', 'Highlight any collaborative research experience, especially with external academic or industry partners.', 'Demonstrate your communication skills by briefly explaining a complex alignment concept in simple terms.']

Research before applying

• Read AE Studio's blog posts or case studies on their AI projects to understand their technical focus and culture.
• Look at the team's LinkedIn profiles to see their backgrounds and research interests, especially any alignment-related work.
• Check if AE Studio has published any alignment research or white papers; if so, study them thoroughly.
• Understand the company's client work in ML/data science to see how alignment research might integrate with their services.

Likely interview topics

Based on the job description, expect questions about:

• Walk me through a recent alignment experiment you designed: what was the hypothesis, how did you test it, and what did you learn?
• How do you approach the problem of interpretability in large models? Can you compare methods like activation patching vs. probing?
• What are the current limitations of pretraining for alignment, and how would you propose to address them?
• Describe a time you had to pivot your research direction based on negative results. How did you decide what to do next?
• How do you stay updated on the latest alignment research, and which recent paper do you think is most impactful?

Common mistakes to avoid

• Don't focus only on generic ML achievements (e.g., Kaggle competitions) without connecting them to alignment.
• Avoid being vague about your research interests; be specific about alignment subfields you're passionate about.
• Don't neglect to show your experimental design skills; alignment research requires rigorous hypothesis testing, not just model building.