Research

Muhammad Zane · AI safety, alignment, and interpretability.

Research

2026 Working paper
Do internal signatures predict how models play?
2026 · Working paper
LMI
2026 Proposed method
What evidence would show a model represents evaluation as an internal state?
2026 · Proposed method
SiteLMI
2026 Working paper
Separating persistence caused by a model from persistence caused by the world around it, with falsification criteria.
2026 · Working paper
SiteLMI

Essays

2026 Essay
Does the present exist in the universe?
2026 · Essay
Read
2026 Essay
How lenders might price debt collateralised by GPU fleets.
2026 · Essay
Read
2026 Essay
What a functioning forward market for GPU compute would look like.
2026 · Essay
Read

Instruments

2026 Interactive
A working GPT-2 in the browser: watch computation move through embeddings, attention, MLPs, and the softmax.
2026 · Interactive
Launch
2026 Field map
The literature as one navigable document: dependency graph, reading pathway, runnable protocols, open problems.
2026 · Field map
Open

In preparation

2026 Safety
Safeguard Durability in Open Weight Models
2026 Cognition
Preemptive Detection of Agentic Misalignment, and Its Shelf Life
2026 Safety
Biosafety Red Teaming of Evo 2
2026 Interp
Biological Theories of Consciousness, Mapped onto Language Model Mechanisms
2026 Cognition
Distress, Flourishing, and Valence in Model Output
2026 Cognition
Introspection and the Reliability of Model Self Report

Collaboration

I work on safety, alignment, evaluation awareness, and interpretability, and I am glad to collaborate. Write to m@latentmindsinstitute.com.