Claude Opus 4.1 has a 50% time horizon of around 1 hr 45 min
(2025)
METR post
Claude Opus 4.1 has a 50% time horizon of around 1 hr 45 min on METR's agentic multi-step software engineering tasks. This is 30% longer than Claude Opus 4, a rate of improvement that matches the observed longer-term trend of around 200 days per time horizon doubling.
CoT May Be Highly Informative Despite “Unfaithfulness”
(2025)
METR post
Recent work demonstrates that LLMs’ chains of thought aren’t always “faithful”: they don’t contain an accurate representation of every cognitive step the LLM used to arrive at its answer. However, perfect faithfulness may be an unnecessarily high standard for practical safety analysis.
As a research engineer contracted with METR, I helped run experiments examining whether harmful reasoning patterns remain detectable despite chain-of-thought unfaithfulness.
The Rethink Priorities Existential Security Team's Strategy for 2023
(2023)
EA Forum
I wrote a report in May 2023 on XST's strategy for this year. We are focused on helping launch entrepreneurial projects that aim to reduce existential risk.
Nanotechnology strategy research resources database
(2022)
Resource database
Here is a database with descriptions and links to over 40 resources relevant for nanotechnology strategy research.
My thoughts on nanotechnology strategy research as an EA cause area
(2022)
EA Forum
Advanced nanotechnology might arrive in the next couple of decades and could have very positive or very negative implications for existential risk. There has been relatively little high-quality thinking on how to make the arrival of advanced nanotechnology go well, and I think there should be more work in this area.
A dataset of non-pharmaceutical interventions on SARS-CoV-2 in Europe
(2022)
Scientific Data
This paper describes a new dataset designed for the accurate recording of NPIs in Europe’s second wave to allow precise modelling of NPI effectiveness. The dataset has considerable potential for use in disentangling the effectiveness of NPIs and comparing the impact of interventions across different phases of the pandemic.
Understanding the effectiveness of government interventions against the resurgence of COVID-19 in Europe
(2021)
(open access)
Nature Communications
We found that business closures, educational institution closures, and gathering bans reduced transmission in Europe's second wave of COVID-19, but less so than in the first wave.
Investigating how technology-focused academic fields become self-sustaining
(2021)
Report
We tracked the time evolution of 29 “factors” for 10 fields. The following factors tend to switch on in the run-up to field establishment: risk tolerant funding, high perceived impact, existence of practical designs, practical work, a high profile paper, and a sense of possibility. Meanwhile, the following factors seem to be unnecessary: a prestigious advocate, intuitive appeal, and strong government interest.
Global Catastrophic Risks 2021: Navigating the Complex Intersections
(2021)
Global Challenges Foundation annual report
I contributed a text on risks and opportunities from advanced nanotechnology to the Global Challenges Foundation's annual report on global catastrophic risks.
Mass mask-wearing notably reduces COVID-19 transmission
(2021)
medRxiv preprint
Our analyses gave a median reduction of 25.8% to the R-number when an entire population wears masks in public.
'Existential Risk and Growth' Deep Dive #2 - A Critical Look at Model Conclusions
(2020)
Report
I investigated how seriously we should take the conclusions from Aschenbrenner's working paper Existential risk and growth.
Coarse-grained modelling of the structural properties of DNA origami
(2019)
(open access)
Nucleic Acids Research
We used a coarse-grained DNA model (oxDNA) to characterise the detailed structural properties of DNA origami structures. The model gave an excellent fit for a particular orgami for which we have high quality experimental structural data.
Simulating large DNA nanostructures with a coarse-grained model
(2016)
Doctoral thesis
I investigated the self-assembly and structural properties of DNA origamis, which are large-scale DNA nanostructures comprising thousands of DNA base pairs, using oxDNA.
Direct Simulation of the Self-Assembly of a Small DNA Origami
(2016)
ACS Nano
We used oxDNA to simulate the complete assembly of a small-scale DNA origami, finding evidence of kinetic traps and co-operativity between staple strands. The paper was highlighted in the "In Nano" article in that journal issue.
Introducing improved structural properties and salt dependence into a coarse-grained model of DNA
(2015)
(open access)
The Journal of Chemical Physics
We introduced an extended version of oxDNA, a coarse-grained model of DNA. This was a featured article and one of the journal's most-read articles in 2015.
Coarse-graining DNA for simulations of DNA nanotechnology
(2013)
(open access)
Physical Chemistry Chemical Physics
We presented an overview of the approaches to coarse-grained modelling of physical processes involving DNA, with a focus on the coarse-grained model called oxDNA.