Project Tips

Project Tips

OVERVIEW

We strongly recommend that all applicants read the details that follow carefully and submit applications that are aligned with our preferences.

We strongly recommend that all applicants read the details that follow carefully and submit applications that are aligned with our preferences.

We strongly recommend that all applicants read the details that follow carefully and submit applications that are aligned with our preferences.

We strongly recommend that all applicants read the details that follow carefully and submit applications that are aligned with our preferences.

  1. Governance Track

Areas of Interest

This track focuses on AI policy, regulation, and related legal frameworks. Our areas of interest this year include:

This track focuses on AI policy, regulation, and related legal frameworks. Our areas of interest this year include:

This track focuses on AI policy, regulation, and related legal frameworks. Our areas of interest this year include:

This track focuses on AI policy, regulation, and related legal frameworks. Our areas of interest this year include:

This track focuses on AI policy, regulation, and related legal frameworks. Our areas of interest this year include:

African AI policymaking

African AI policymaking

African AI policymaking

African AI policymaking

African AI policymaking

EU AI law and Global South AI safety

EU AI law and Global South AI safety

EU AI law and Global South AI safety

EU AI law and Global South AI safety

EU AI law and Global South AI safety

EU AI law and Global South AI safety

Whistleblower protection in AI safety

Whistleblower protection in AI safety

Whistleblower protection in AI safety

Whistleblower protection in AI safety

Whistleblower protection in AI safety

Whistleblower protection in AI safety

Sociotechnical AI misuse evaluations

Sociotechnical AI misuse evaluations

Sociotechnical AI misuse evaluations

Sociotechnical AI misuse evaluations

Sociotechnical AI misuse evaluations

Sociotechnical AI misuse evaluations

Threat modelling for Global South contexts

Threat modelling for Global South contexts

Threat modelling for Global South contexts

Threat modelling for Global South contexts

Threat modelling for Global South contexts

AI agents regulation

AI agents regulation

AI agents regulation

AI agents regulation

AI agents regulation

AI agents regulation

Legal doctrines relating to AI safety

Legal doctrines relating to AI safety

Legal doctrines relating to AI safety

Legal doctrines relating to AI safety

Legal doctrines relating to AI safety

Legal doctrines relating to AI safety

Read detailed descriptions here.

Read detailed descriptions here.

Read detailed descriptions here.

Read detailed descriptions here.

Read detailed descriptions here.

Governance Track Project Design

We welcome projects aimed at producing publishable outputs, such as research reports, journal articles, or substantial blog posts. The most competitive applications propose projects with research questions that are clear, specific, and demonstrate a distinct contribution relative to existing work.

Writing a Compelling Governance Track Proposal

We recommend that governance track applicants structure their proposal in three parts as follows:

Part 1 should be a section where:

(i) You list your main research question and 2 to 4 underlying research questions (which feed into your main research question) that your project aims to explore.

(ii) You discuss why you concluded that it’s important to study your main research question.

Part 2 should be a section where:

(i) You list the hypothesis of your study (In other words, what you intend to say in response to your main research question).

(ii) You explain how your study will be unique in comparison to other work that exists on the same subject. For example, you could show this by demonstrating that no one has done comprehensive research on your main research question yet; or your methodology is different from those used before; or your hypothesis is original. 

Part 3 should be a section where:

You explain – as specifically as possible – the methodology you intend to use to carry out your study. That is, for each underlying research question, which sources of data will you rely on, and how will you analyze them to extract the information that you need? You should explain this in a detailed manner in your own words (we’re not interested in whether or not you’ve used the most recognized scientific term for a method). 

Please note that proposals are more likely to be successful when the main research question is sufficiently clear and specific to be answered within the duration of the fellowship.

  1. Technical Track

Areas of Interest

This track focuses on AI safety evaluation and mechanistic interpretability. We are particularly interested in research involving:

This track focuses on AI safety evaluation and mechanistic interpretability. We are particularly interested in research involving:

This track focuses on AI safety evaluation and mechanistic interpretability. We are particularly interested in research involving:

This track focuses on AI safety evaluation and mechanistic interpretability. We are particularly interested in research involving:

This track focuses on AI safety evaluation and mechanistic interpretability. We are particularly interested in research involving:

Designing reproducible evaluation pipelines

Designing reproducible evaluation pipelines

Designing reproducible evaluation pipelines

Designing reproducible evaluation pipelines

Designing reproducible evaluation pipelines

Building automated red-teaming systems

Building automated red-teaming systems

Building automated red-teaming systems

Building automated red-teaming systems

Building automated red-teaming systems

Building automated red-teaming systems

Analysing AI representations and activations

Analysing AI representations and activations

Analysing AI representations and activations

Analysing AI representations and activations

Analysing AI representations and activations

Analysing AI representations and activations

Prototyping safety interventions (fine-tuning, runtime filters, activation patches)

Prototyping safety interventions (fine-tuning, runtime filters, activation patches)

Prototyping safety interventions (fine-tuning, runtime filters, activation patches)

Prototyping safety interventions (fine-tuning, runtime filters, activation patches)

Prototyping safety interventions (fine-tuning, runtime filters, activation patches)

Prototyping safety interventions (fine-tuning, runtime filters, activation patches)

Read a more detailed description here.

Read a more detailed description here.

Read a more detailed description here.

Read a more detailed description here.

Read a more detailed description here.