Simulating Neurological Disorders in LLM Weights

Reproducible, ethical testbed for Alzheimer's, Parkinson's, and other neurological disorders using mechanistic interpretability.

InfiniteZero Network — Global AI Commons Infrastructure

Trustless, open source protocol for collective AI training where data never leaves the device.

Stress-testing alignment auditing methods of AuditBench

Empathic Transaction Notation: A Formal Framework for LLM Red-Teaming

Formal notation and synthetic dialogue pipeline for predicting escalation targets in empathic Human-LLM interactions

LLMMatrix: Synthetic World Benchmark for Contamination Free LLM Forcasting

A contamination-free benchmark for measuring whether LLMs can forecast, or whether they're just remembering.

Φ-Arena and adversarial multi-agent VLA interpretability

Independent collective. Φ-Arena open benchmark, 3 ICLR 2027 papers (Φ-Arena, mechinterp, energy-bounded) — kickstart for a 10-year program.

When Language Becomes Workflow

Provenance Failures in Tool-Bearing LLM Agents

Harmonic: Hierarchical SSM for Long-Context Language Modeling

MATS 10.0 Neel Subliminal Learning

Awareness-conditioned emergent misalignment

Does harmful fine-tuning data cause broad misalignment only when the model already recognises the target behaviour as a norm violation?

Travel to the UN Global Dialogue on Artificial Intelligence Governance in Geneva

I've been invited as I have contributed with input to prepare the meeting.

WorkstationLLM: defensive primitives toolkit for local-AI operators

Three Apache-2.0 daemons (memory guard, request queue proxy, disk watchdog) that kept my 24-app local AI stack crash-free for 6+ months.

Token-level interpretability of long-chain reasoning in transformer models

Local-first experiments on reasoning trajectories, with cluster/API validation and ICLR 2027-oriented outputs.

Alignment System to Replace RLHF

Triadic geometric training data and architecture replaces RLHF

Travel Grants for New Zealand's Inaugural AI Safety Conference

$5,000 USD to bring 16 vetted academics and policy leads to NZ's first AI Safety Conference, ensuring national representation and cross-sector collaboration

Falsifiable Mechanisms of AI Oversight Erosion via Recursive Displacement

33 named mechanisms mapping how AI deployment degrades human oversight. Public API + MCP tool. Competence Insolvency corroborated independently by Karpathy.

Stratum: A Programming Language with Structural AI Safety Properties

Programming language that enforces AI safety as runtime substrate properties rather than training-time alignment, already shipping on real hardware.

A Small AI Governance Evidence Kit from Türkiye

A micro-project to turn an early AI governance prototype into risk notes, oversight checklist, review templates and public learning.

Human-Side Alignment Risks in Non-Self Language Models

Projection, sycophancy, and institutional artifact fabrication in AI-mediated supervision.

Lyptus Research: Bridge Funding

An early-stage AI safety research group based in Sydney, Australia