Stephen Elliott

Stephen Elliott

Independent Researcher and Convenor of Gigascale Labs

Sitemap

A list of all the posts and pages found on the site. For you robots out there, there is an XML version available for digesting as well.

Pages

Posts

Paper Notes: Emergent Cooperation and Strategy Adaptation in Multi-Agent Systems: An Extended Coevolutionary Theory with LLMs (Zarzà et al., 2023)

4 minute read

Published:

Welcome to Paper Notes, where we record our groups’ weekly discussions of innovative papers from across artificial intelligence. On Tuesday the 21st of October, our mathematics reading group read Emergent Cooperation and Strategy Adaptation in Multi-Agent Systems: An Extended Coevolutionary Theory with LLMs, published in MDPI Electronics in 2023. The authors come from a collection of three universities and a lab, across Spain and Germany.

Paper notes: A Practical Review of Mechanistic Interpretability for Transformer-Based Language Models (Rai et al., 2025)

1 minute read

Published:

On Friday the 5th of September, the general reading group continued its mechanistic interpretability sprint with A Practical Review of Mechanistic Interpretability for Transformer-Based Language Models. The research team comes from across several US universities, with one member from Salesforce Research. Its lead author is a PhD student, and its second author a PhD working in industry. The survey was posted on arXiv and presented as a tutorial at ICML 2025.

Paper notes: On the Biology of a Large Language Model (Lindsey et al., 2025)

2 minute read

Published:

Last week, Deep Network’s reading group read On the Biology of a Large Language Model. The research team comes from Anthropic’s interpretability research group, and was published in the Transformer Circuits interactive research thread as well as on the Anthropic website.

projects

How Machine Learning Extends Software Engineering

Published:

A bank transfer system. A stock exchange. A physics simulator. These are classic examples of traditional software engineering: deterministic, rule-based instruction sets written by humans. Machine learning extends traditional software engineering by applying statistics to software decision making. With statistics, software can learn from observations of the world, generating the instruction sets autonomously.

Deep Network Permalink

Published:

UNSW AI reading group I ran, which covered foundational and frontier research in capabilities, interpretability, and multi-agent systems. Special activities included our YouTube, a research presentation night, and hosting a Sydney Hub for the Apart AI Control Hackathon.

Engineering Catalyst, A Scalable Machine Learning Framework

Published:

I wrote a term paper for my Advanced Algorithms course at UNSW, covering my software engineering work on a deep learning framework I call Catalyst. The report covers the software engineering theory behind deep learning and concurrent/parallel/distributed theory, with attention to both my Catalyst implementation and industrial frameworks in general.

Download Paper

Founding and Engineering UniMate

Published:

I was a co-founder and sole engineer for UniMate in the second year of my degree. It was an event aggregation and data processing system that integrated distributed proxies for web scraping, GPT-based multi-label classification, modular I/O architecture, CLI interaction, dynamic HTML parsing, and a failure-tolerant data pipeline with partial restart. It ran on a no-code frontend. This was my first real-world system, and it taught me about design, trading off speed and quality, and minimal engineering for scalability.

My Paper Reviews For 2025

Published:

Alongside Deep Network’s meetings in 2025, our members wrote paper reviews, including reflections on the papers ideas and analysis of research standards.

A Safety Evaluation Of Moltbook Posts

Published:

Moltbook is the first big, organic society of advanced artificial intelligences. It provides a valuable early dataset for understanding large multi-agent system risks.