Hypothesis Prism: A Structural Representation Layer for Scientific Experiments

LLMs scale language, not method. HP turns hypothesis-driven papers into machine-readable maps of variables, controls, stats, and findings for researchers & AI.

Stipend to build formal proof training data for mathematical AI safety

Training graduate mathematicians as Lean formalisers in Germany to produce human-verified proof data that LRMs urgently need

SOPHIA: Open-Source Epistemic Constitution for AI Reasoning

An executable reasoning quality framework that checks whether AI-generated arguments are logically sound — not just factually accurate. Live at usesophia.app.

Safe AI Germany (SAIGE)

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

Distinguishing Intelligence from Agency in Frontier AI Systems

Empirical testing of whether AI capability scaling leads to emergent agency or shutdown resistance in frontier systems.

Job Security Compass

An Interactive Tool for Navigating AI Career Risk

Eliminating the Effective Altruism (EA) Grant Evaluation Bottleneck with Graev

Seeking funding to secure API infrastructure and permanently eliminate the rate limits bottlenecking open-source EA grant evaluation.

Help a Bootstrapped AI Risk Literacy Founder Get To IASEAI 2026 in Paris

Nutrition labels transformed food safety through informed consumer choice, help me do the same for AI and make this standard :)

Towards Shared AI Compute Infrastructure

Law School for AI Governance

Support my postgraduate law studies and research in AI Governance

Real-Time Detection of Hallucinations in Large Language Models

Measuring and Visualizing Model Uncertainty During Inference

AI Security Startup Accelerator Batch #2

Funding 5–10 AI security startups through Seldon’s second SF cohort.

[Retroactive] Funding for developing new "substrate-flexible risk" threat model

I self-funded research into a new threat model. It is demonstrating impact (accepted at multiple venues, added to BlueDot's curriculum).

Formal Certification Technologies for AI Safety

Developing the software infrastructure to make AI systems safe, with formal guarantees

Fund thebes' model tinkering

Research, tutorial writing, and open-source libraries & tools for experimenting with language models

FloraForge: Aesthetic Gene Editing in Petunias

A modular gene-editing platform for engineering new pigment patterns in ornamental plants, starting with a vein-pattern rescue line in petunias.

Software to detect data fabrication

Continue developing ‘copy-paste-detective’ - a software that detects signs of data fabrication in scientific research - and run it against 20,000 publicly available Excel datasets.

SDCPNs for AI Safety

Developing correct-by-construction world models for verification of frontier AI

TAAC: Cross-Provider Safety Analysis of Prompt Compression in LLM Inference

Novel safety failure modes discovered across 7 LLM providers with 35,000+ controlled inference trials. Targeting NeurIPS 2026.

Neuro-Safety: Reducing AI Latency for Bio-Safe Interaction

Developing an Edge-AI framework to reduce response latency to <0.6s, mitigating user cognitive stress and establishing "Digital Pharmacotherapy" standards.