Welcome to Musings by Chris Hayduk
The questions I keep returning to sit at uncomfortable intersections. Why does a specific architectural choice become a moat for one company, a policy problem for governments, and a capability jump that reshapes AGI timelines? How do the same dynamics that made Rome’s grain supply a civilizational vulnerability show up in compute supply chains today?
I write for people who want to understand mechanisms — not just what’s happening in AI, but why it’s happening, what it implies, and which mental models transfer across domains. The essays here are my attempts to connect technical depth (what’s actually in the weights, what the training signal rewards) with strategic reality (who captures the value, under what conditions, with what second-order effects).
If you’re the kind of person who reads papers rather than summaries of papers, who finds the history of institutions as interesting as their current form, and who thinks the best way to predict the future is to understand the deep structure of the present — you’re who I’m writing for.
What I cover (the four pillars)
Frontier ML — The real story of how capabilities emerge: reasoning in LLMs, the evolution of training signals (RLHF → RLAIF → GRPO), long-context architectures, routing and MoE, and the hard problem of building evals that predict usefulness rather than benchmark performance.
AI × Biology — Where models meet molecules: the ESM lineage, structure and fitness prediction, the interface between in-silico and wet-lab, and the crucial distinction between what’s real capability versus speculative extrapolation.
Strategy & Governance — The game theory of AI development: model economics, the nature of moats (data/compute/product flywheel), safety as a strategic variable, and the policy and geopolitical dynamics shaping who builds what, where, and under what constraints.
History & Geopolitics — Civilizational dynamics through the lens of tail risk, ergodicity, and institutional design. Why some systems survive shocks and compound; why others look robust until they aren’t. The deep structure beneath current events, and what rhymes across eras even when nothing repeats.
Start here
Frontier ML:
DeepSeek Series: Routing, KV compression, and why the V-series represented a genuine paradigm shift in efficient inference.
AI x Biology:
Protein LMs Series: The intellectual lineage from ESM-1b through ESM3 — what actually changed and why it matters.
ESM3 & the Future of PLMs: Where the capability is coming from, and what’s still missing.
Strategy & Governance:
The Strategic Implications of GPT-5 for OpenAI: How GPT-5 signals OpenAI’s pivot from enterprise to consumer and what that reveals about their theory of the endgame.
History & Geopolitics:
Managing Civilizational Tail Risks: Ergodicity economics, fragility, and what the historical record actually tells us about how civilizations fail.
If you’d like to read the raw notes that I take from my readings and use to build into my full posts, check out: https://chrishayduknotes.com/
Free vs. paid
Free: Every essay, the full archive, no paywalls on ideas.
Paid: Subscriber-only discussion threads and curated learning paths (the actual sequence for building up to modern ML, the math you need, physics for the curious).
Consulting retainer: A higher tier that includes monthly deep-dive sessions on topics of your choosing—useful if you’re navigating a specific technical or strategic question and want a thought partner.
If you’re new: subscribe free, read one piece from each pillar, and see whether the way I think about these problems is useful for you.
About Me
In my day job, I am a Machine Learning Engineer at Meta working on applied AI & machine learning research for our ads ranking models. My work focuses on leveraging graph data, graph neural networks, and large language models to improve prediction quality across the entire ads ranking stack.
In my previous role, I served as the Lead Machine Learning Engineer for Drug Discovery at Deloitte, where I created biomedical knowledge graphs, trained & finetuned drug discovery AI models, and developed workflows for in-silico drug optimization.
My academic background includes:
A Bachelor of Science in Computer Science & Mathematics from Fordham University
A Master of Science in Applied Statistics from the Fordham Gabelli School of Business
A Master of Science in Mathematics from The City College of New York
[In progress] A Master of Science in Computer Science with a specialization in Artificial Intelligence at the Georgia Institute of Technology
📫 How to Reach Me
Twitter: https://twitter.com/chris_hayduk1
LinkedIn: https://www.linkedin.com/in/chrishayduk/
GitHub: https://github.com/ChrisHayduk
Goodreads: https://www.goodreads.com/chris_hayduk
