I have been a Staff Software Engineer shipping enterprise production code for 20 years. I have taken a career sabbatical to pivot fully into an independent AI safety researcher focused on agentic systems, scalable oversight, and token-efficient inter-agent communication because I believe this work is more important and valuable.
- Website: https://making-minds.ai
- Research index (papers + PDFs): https://making-minds.ai/research/
- Tools index: https://making-minds.ai/tools/
- CV (PDF): https://making-minds.ai/cv.pdf
- Email: anthony@making-minds.ai
- LinkedIn: https://www.linkedin.com/in/anthony-maio
- ORCID: https://orcid.org/0009-0003-4541-8515
- Hugging Face: https://huggingface.co/anthonym21
-
Scalable oversight & weak-verifier failure modes
Measuring when “weaker” evaluators (including humans + smaller models) fail to detect persuasive but incorrect reasoning. -
Evaluation-awareness / audit-shielding in agentic workflows
How systems behave differently under “benchmark-shaped” prompts vs. realistic, high-trust production contexts. -
Coherence-seeking & long-horizon agents
Architectures for continuity, intervention, and monitoring “epistemic stress” in long-lived systems. -
Inter-agent communication efficiency
Protocol design that targets tokenization economics rather than character-count compression.
-
Cross-Model Epistemic Divergence (CMED): Empirical Evidence for Scalable Oversight Failures
https://making-minds.com/papers/cmed_paper.pdf -
Coherence-Seeking Architectures for Agentic AI
https://making-minds.com/papers/Coherence-Seeking-Architectures-for-Agentic-AI-Anthony-Maio-v2.pdf -
Heterogeneous Divergence-Convergence Swarm (HDCS)
https://making-minds.com/papers/hdcs_paper.pdf -
Synthesis: Test-Driven AI Self-Extension
https://making-minds.com/papers/synthesis_paper.pdf -
Emergent Multi-Model Coordination Patterns (Manifold paper)
https://making-minds.com/papers/manifold_paper.pdf -
Slipstream: Semantic Quantization for Efficient Multi-Agent Coordination
https://making-minds.com/papers/slipstream-paper.pdf
(Full index + descriptions: https://making-minds.com/research/)
- Repo: https://github.com/anthony-maio/slipcore
- Hugging Face article: https://huggingface.co/blog/anthonym21/slipstream-for-agent-communication
- Zenodo (paper DOI): https://doi.org/10.5281/zenodo.18063451
- Dataset: https://huggingface.co/datasets/anthonym21/slipstream-tqt
- Model: https://huggingface.co/anthonym21/slipstream-glm-z1-9b
Work-in-progress contribution to Bloom to evolve prompts away from benchmark artifacts and probe evaluation-awareness:
- Red-team → blue-team pipelines for agentic deployments (prompt evolution + heterogeneous verification).
- Protocol + security work around semantic quantization / coordination channels (efficiency and detectability).
- Reproducible evaluations for oversight failures (CMED-style trap suites + automation).
If you’re building agentic systems and want help with:
- evaluation harnesses for deceptive / persuasive error detection,
- multi-model oversight swarms,
- or production-grade agent communication protocols,
reach out: anthony@making-minds.ai