Independent research lab.

Formal foundations of AI reasoning — spanning computational complexity, verification theory, and large-scale empirical evaluation. Based in Reykjavik.

Research

On the Reasoning Gaps of Large Language Models: A Formal Characterization

Pre-Print

NeurIPS 2026

176,000 evaluations across 12 models. Systematic failure modes correlate with computational complexity classes.

The Computational Complexity of Verifying LLM Outputs

In Progress

ICLR 2027

Formal framework for verification efficiency. When is checking harder than generating?

A Taxonomy of Failure Modes in LLM-Based Autonomous Agents

In Progress

ACL 2027

50+ deployment incidents analyzed. Structured taxonomy of agent failure modes.

Impossibility Results for Unsupervised Self-Improvement in Language Models

Early Stage

ICLR 2027

Theoretical bounds on what self-learning can and cannot achieve.

All papers →

About

Serre AI is a solo research operation by Oddur Sigurdsson. The lab uses autonomous AI agents to conduct research 24/7 — reading literature, designing experiments, writing papers, and iterating on their own weaknesses.