Projects
Recent academic projects and papers.
- Measuring Agent Capabilities on Open-ended AI R&D Tasks
Tracking frontier model progress on four long, open-ended AI R&D tasks.
- A Natural Deduction Theorem Prover for Large-scale Expert Systems (treesolve)
Funded as part of ARIA's Safeguarded AI workstream, a SAT-based theorem prover generating human-readable proofs.
GitHub(Paper under review)
- A Modular ASPIC+ Argumentation System (pickaxe)
A high-performance, modular, and pure-Rust implementation of the ASPIC+ framework (Modgil & Prakken, 2013).
GitHub(Paper in preparation)
- Evaluating Language Model Character Traits
An empirical, behaviourist account of the character traits exhibited by large language models.
- The Propensity for Density in Feed-forward Models
A study evaluating the sparsity of deep learning networks.
- Finding Sparse Initialisations using Neuroevolutionary Ticket Search (NeTS)
An evolutionary algorithm for finding efficient subnetworks in feed-forward and convolutional DNN architectures.
