Research
Papers
Sep 22, 2026
Attention as Routing Graph: Circuit Extraction from Single Forward Pass
Treat attention as a routing map from one forward pass, keep a small subgraph aimed at the answer, and check it under ablation. On induction and IOI across GPT-2 Small/Medium and Pythia-410M, extracted edges hurt far more than random sets of the same size. arXiv:2609.25285.
Sep 8, 2026
Finding Usable Weight Mechanisms with Tiled SVD
We extract mechanism mounts directly from weight sites by column-tiled SVD — each mount a triple (v, u, {σ}) as trigger, write, and strength. Pre-registered A/B/C eval on Gemma-2-2B (WikiText-2): 182/182 GO. arXiv:2608.06969.
Applied Research
Oct 26, 2026
Replication of Single-Nucleus Transcriptomic Findings
Cross-study replication of snRNA-seq findings. Uniform pipeline, replication table, CSF/plasma biomarker shortlist. Target December 2026.
Oct 20, 2026
axi
Auto RL self-learning agent for controlling developer environments like aq at full efficiency: coding and research with harness lessons, rejected paths, and refine loops.
Oct 14, 2026
aq
Developer environment for foundation models: frameworks, libraries, tools, and agents to design, test, and refine models with precision and intent.
Oct 6, 2026
World Models
Talking-avatar LoRA on MiniMax-H3, an action-conditioned JEPA for Mario with probe gates, and a 21MB pixel-art world model.
Sep 28, 2026
Interpretability
Applied research on AQIT: transformers and LLMs, attribution, sparse autoencoders, simulation, live training watch, embedding models, evals, security, and benchmarks. Full original writeups in one article.
Sep 16, 2026
Experimental Weight Editor
Agentic ROME on Pythia 2.8B: causal trace layer location, rank-one MLP updates, and a three-check validation loop that rolls back and retries on failure. Includes case studies on factuality, bias correction, and censor auditing.
Sep 2, 2026
Simulating LLM Behaviour in Different Environments
Four environments for studying LLM behaviour: activation steering for bias reduction, Chess Agents Simulation, Echoes gossip RPG, and Among Us Agents Simulation.
