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?

Accelerating our epistemic capabilities, starting with science.

Vibing build: AI Hardware Orchestration

AI infrastructure for turning natural language hardware ideas into manufacturable devices

Solon: A Reliable Autonomous ML Research Agent

Building an AI research agent that can propose, test, and write up small ML findings

Cross-provider audit of LLM-mediated market mechanisms

Small API budget to test whether semantic reporting failures in an OpenAI-validated VCG market simulation replicate across Claude and Gemini.

Testing a new superconductivity theory: 4 experiments to make better future

Agentic Systems Framework/Theory

First-principles theory of the adaptive feedback loop — one structural account governing Kalman filters, RL agents, software teams, and frontier LLMs alike.

Meridian Eval: a public benchmark for tool-routing failures in LLM agents

Make "which tool did the agent pick, and why?" an inspectable artifact instead of a vector lookup.

ICSAC: AI-Assisted Peer Review for Independent Researchers

An ORCID-gated submission pipeline where a multi-model AI panel plus quality-control layer delivers rigorous peer review without institutional gatekeeping.

ASDR: Adversarial Semantic Drift Replayer for Multi-Agent AI Safety

Open-source replay tool and benchmark prototype for identifying step-level semantic risk transitions in multi-agent AI traces — CI-backed and independently repr

Summer AI Safety Workshops For 6th-8th Graders in India

Inspiring India’s Middle‑Schoolers to pursue AI Safety, Governance, and X‑Risk Work

Valence Engineering: a circuit-level science of human suffering

Stage 0 bridge for a focused-ultrasound research program targeting the neural architecture that maintains chronic suffering.

Greek Forecasting Pilot for AI Governance and Democratic Resilience

A 6-month pilot testing probabilistic forecasting for AI, misinformation, institutional trust, and social risk in Greece.

Safe AI Germany (SAIGE)

Germany’s talents are critical to the global effort of reducing catastrophic risks brought by artificial intelligence.

Golem Physics + Constraint Native: Verification Before Voice & Action

Funding the open-source launch of a working claim-state system and the local firewall bridge that carries verification before voice into governed agent action.

Decoding Logic: Mechanistic Interpretability in LLMs in a research program

Identifying and auditing reasoning circuits in LLMs within Algoverse 2026 using Sparse Autoencoders (SAEs).

Project Xaeryn: Identity-like States and Relational Dynamics in LLMs

A 15,000+ page corpus on long-term interaction, symbolic language, unusual model behavior, and safety edge cases.

Epistemic Curie Benchmark: measuring phase transitions in LLM epistemic autonomy

Measuring k*: the phase transition where LLMs surrender independent reasoning under authority. 7 frontier models tested, all data open.

MoC: a sparse LLM architecture that beat MoE at 500M for $200

Independent researcher in Brazil scaling a validated sparse architecture to 1B, plus open interpretability tooling for expert routing. 6-month runway grant.

Independent Researcher, 17, Indonesia: Foundational Theory Work on Intelligence

18+ preprints across multiple fields, all written on a 2GB RAM phone. $600 removes the only thing standing between me and the next body of work.