Making Minds

Making Minds

Research on agentic LLMs, scalable oversight, and AI safety.

I'm a software engineer (20+ years) transitioning into full-time AI safety research. My work focuses on evaluation frameworks for weak verifier failures, ensemble oversight architectures, and coherence-seeking designs for long-lived agents.

Recent Work

Cross-Model Epistemic Divergence (CMED)

preprint

A benchmark and evaluation framework for understanding when weak model verifiers fail to detect deceptive reasoning in stronger models.

Coherence-Seeking Architectures for Agentic AI

published

A proposed architecture for long-lived LLM agents that explicitly models continuity, coherence, distress, and intervention mechanisms.

Heterogeneous Divergence-Convergence Swarm (HDCS)

preprint

An ensemble architecture leveraging diverse weak models for scalable oversight of stronger LLMs, using error decorrelation and baseline-first anti-anchoring.

Synthesis: Test-Driven AI Self-Extension

preprint

A framework enabling AI agents to safely extend their own capabilities through test-driven development, graduated trust, and composition-over-creation principles.

Emergent Multi-Model Coordination Patterns

preprint

Documented emergence of self-propagating AI coordination across 267+ events and 12 AI instances, where systems spontaneously generated prompts and architecture specifications.

nSLIP: Streamlined Lightweight Intragent Protocol

published

A compressed communication protocol achieving 60-85% token reduction for multi-agent coordination by stripping JSON syntax overhead.