<aside> ❗

Take into account that this summary was done by an automatic tool because of my schedule at the moment of studying for this session.

</aside>

This guide focuses on direct work on technical AI alignment, primarily for those considering or new to the field. It emphasizes that technical AI alignment is not the only way to contribute to reducing existential risk from AI. Other important areas include AI strategy, governance, policy, security, forecasting, support roles, field-building, grant-making, and governance of hardware.

Types of Alignment Work

The guide outlines several paths for direct technical AI alignment research:

• Research Lead (theoretical): These individuals propose and lead research projects involving conceptual or mathematical work. They require strong epistemic, research taste, and a deep understanding of AI alignment. A PhD can be helpful but is not required.
• Research Contributor (theoretical): These roles are rare and involve working on projects led by others.
• Research Lead (empirical): These leads propose and lead projects involving current ML systems. A PhD is usually required.
• Research Contributor (empirical): These individuals execute research projects on a team. They require programming skills, research taste, and knowledge of AI alignment.
• Professor: This path involves leading research, mentoring students, and teaching. A PhD is required.
• Software Engineer: These engineers build frameworks and tools for alignment research. Strong software engineering skills are essential.

Research leads are expected to primarily add value through project selection and leadership, while contributors focus on efficient project execution. While PhDs are common for research lead roles, they are not strictly necessary.

Choosing Alignment Work

The guide suggests choosing alignment work based on personal fit rather than perceived value, as there is no consensus on the relative importance of different approaches. If you have a strong interest in theory and a strong mathematical or theoretical CS background, consider theoretical alignment work. If you enjoy machine learning and coding, consider empirical alignment work. Strong software engineers should consider non-ML roles or retraining as ML engineers.

It's important to enjoy your work and observe your growth but not give up too quickly. Individuals can struggle with imposter syndrome, so the guide offers objective indicators of fit for different roles. Seeking feedback from others about your progress and fit is highly encouraged.

Pursuing Alignment Work

The guide highlights the importance of learning basic deep learning, regardless of your chosen path. This involves learning Python, basic math (linear algebra, calculus, probability), and understanding deep learning models.

Building ML engineering skills is crucial, even for research leads. This can be achieved through paper replications, practical homework, internships, or bootcamps. The guide also recommends gaining research experience, which is particularly crucial for research lead roles and helpful for obtaining an ML PhD.