Research

I aim to produce research which meaningfully shifts how the research community, public and policy-makers think about LLMs and AI more generally. My focus is on forward-looking research which studies phenomena which seem likely to occur within the next 1-2 years. I am currently early in my PhD and have not decided on what area to specialise in; I describe my general interests below.

Interactions & multi-agent systems. I am interested in the interfaces, protocols and interactions between agents, whether human or AI. From a human impacts perspective, much of AI psychosis/belief influence/cognitive offloading/etc is driven by how AI systems and humans interact. From a multi-agent system perspective, I’m super interested in the possible consequences of artificial agents interacting, observing and learning from one another.

Trusting LLMs. I trust a tool or person if I believe that they are so likely to perform a task in the manner I intended that I don’t check whether they did so. LLMs are heavily optimised to gain the trust of their users, through performing competence, reflectiveness and objectivity. To a large degree, this trust is what makes LLMs useful: they are able to autonomously exercise judgement about sub-problems encountered when working towards the user’s stated objective. However, as LLM capabilities advance and people and institutions trust LLMs more, they are likely to be provided greater autonomy, less supervision and more powerful tools; and integrated into the social and political processes our societies depend on. I aim to investigate possible consequences and develop mitigations, if possible.

LLM-powered autonomous agents. Trillions of dollars of investment are currently aimed at building autonomous and long-running AI agents. These agents will be both more and less powerful, more and less intelligent, more and less moral than humans, depending on what aspect of each concept one focuses on. They will interact with one another, and also with humans. There are many questions raised by this new paradigm: how agents will influence one another; how systems relying on accountability or reputation apply to agents; and how such agents and humans might work together.

However, my research interests are not set in stone. I’m also interested in monitorability, the exhaustiveness & implications of the persona selection model, and many other topics. If it is novel or forward-looking, addresses an important problem, and is related to AI safety, I am likely interested!

Funding

AI research is unfortunately expensive. Additional funding thus meaningfully increases my ability to conduct impactful and high-quality research. I’m grateful to the following organisations for their support:

2026: Oxford Internet Institute (~£100k)
2026: British Open-Ended Learning & Discovery lab (~£66k)
2026: Lyceum (~£700)
2026: Mercor (~50k USD)

All of my funding sources are disclosed above. External funding does not dictate what topics I pursue.¹

Please reach out if you’d like to collaborate on exciting and impactful research, and/or provide support!

In fact, it often enables me to pursue exciting topics that I’d otherwise be unable to investigate! ↩︎

Publications

The Missing Metrics in DSA Content Moderation Transparency

2026-01 DSA Observatory

The Digital Services Act makes platform transparency reporting mandatory and standardised, but the metrics it requires still fall short of what is needed for real accountability. Counts of removals and appeals alone cannot tell us whether content moderation systems are accurate, proportionate, or effective, making the absence of evaluation metrics such as precision and recall increasingly difficult to justify under the DSA’s risk-based logic. While these metrics are unlikely to surface in baseline transparency reports under Articles 15 and 24, the post argues they may yet emerge through heightened scrutiny of the largest online platforms and search engines (VLOPSEs), as regulatory expectations take shape through enforcement, systemic risk reporting, audits, and related obligations.

Computational Models of Human Decision-Making with Application to the Internet of Everything

2021 IEEE Wireless Communications, 28(1), 152–159

This paper presents computational models of human decision-making and their applications to IoT systems. We explore how understanding human behavioral patterns can inform the design of more effective and user-centered Internet of Everything architectures. Co-authored with Setareh Maghsudi.

Random interests

What causes people to change their beliefs
How people (and/or agents!) model one another’s cognitive processes
Philosophical aspects of language models (see these two articles 1 2)
Forced choice and Bradley-Terry/multinomial logit vs scalar measurements
Extensions of prediction-powered inference
Measuring hard-to-measure concepts
Information flows & decision-making inside organisations & shocks to organisations
Algorithm design (think LeetCode but for fun)
Computational social choice; mechanism design; algorithmic game theory
Network science
Behavioural embeddings for clickstream data
Effective altruism
Quant interview-style brainteasers
What is the possible social justification for credit cards with interest rates as the primary revenue driver (spoiler: I don’t think there is one)