Personal essays on AI and humanity.
Learning notes as I transition into full-time work on AI safety and AI for good — with plans to publish on X and LessWrong.
Essays
biology
Cosmos
AI Companies
-
The Claude Fable Episode: Known Facts, Political Artifacts, and a Trust Crisis
What we can and cannot confirm about Anthropic's June–July 2026 export-control saga; why it is a Winner-style politicized product; and what epistemic risks follow when the process is opaque.
-
Why is switching AI platforms so hard -- Memory
AI platforms are building the deepest lock-in in the history of consumer technology. Switching cost theory, six historical case studies, and Anthropic's memory import experiment explain why preference portability protocols will fail without regulation — and why the window for intervention is closing.
Philosophy
-
Entering the Post-/Transhuman
Paleolithic emotions, medieval institutions, god-like technology. Learning made us Earth's strongest species; our tech is creating something stronger still — so we must upgrade ourselves for tomorrow.
-
Universal values? A social-science and history map
When AI labs cite the UN Declaration of Human Rights, they inherit a century of debate about what 'universal' means — and what it doesn't. Canonical texts, irreconcilable conflicts, cultural variation, and the theories that explain the gaps.
policy
-
What Actually Happens Before Policy Changes
Chernobyl didn't create nuclear regulation from scratch. FTX didn't produce US federal crypto law. A field guide to the years of preparation, policy entrepreneurs, and failed shocks before regulatory punctuation — including the EU AI Act and implications for AI governance.
-
China doesn't care about AI safety? Two different meanings of 'safe'
Second-hand caveat: the US default is that China ignores AI safety; people in Chinese industry say they are doing it quietly. Align the definition first, then look at what is already built, whether top-down is really fast, and what still looks thin.
-
Science, technology, and politics are always intertwined
When people say AI or science is 'political,' they often mean scientists lie or capitalism ruins everything. Langdon Winner's framework is sharper: technical arrangements settle disputes, some technologies favor certain power structures, and science as an institution is never value-free.
AI governance
-
How to make governments care about AI regulation
70% of the public wants AI rules. Almost nothing happens. A political-science playbook — Olson, punctuated equilibrium, policy windows — for moving regulation before a crisis, without waiting for politicians to 'believe' in x-risk.
-
A political map of US AI policy
No federal AI safety law — policy is written in state capitols and fought in primaries. Seven factions, three state templates, SB 53/RAISE in full, federal preemption wars, and the $7M PAC fight over Alex Bores in NY-12.
-
A map of AI governance
Dozens of institutions, summits, and declarations. Almost zero enforcement. For every $1 spent on AI safety, $600-1,200 goes to capability. Here's what actually exists, what power it has, and what's missing.
-
Personal notes: resources for getting started in AI safety
A personally curated starter list — community, training programs, fellowships, job boards, and optional governance engagement. Sorted by barrier to entry; US, China, and international.
-
The happy path with AI
Seven problems we have to face in parallel—status, clocks, success/failure forks, and what is worth doing under real friction: control, bio/misuse, domestic governance, US–China, distribution, who writes the constitution, human capability and meaning.
AI Safety
-
How much does it cost a malicious actor to build powerful AI?
Most harmful use doesn't need a frontier training run. A cost map from $0 jailbreaks to $100B clusters — plus what FBI complaint data actually shows about AI crime today.
-
What are we aligning to? A map of alignment paradigms
RLHF, Constitutional AI, CIRL, CEV, oracle-only Scientist AI, and Gabriel's fair-treatment-of-claims framework are not interchangeable fixes for the same problem. Each silently picks a different answer to what human values are — and most of the field never states which answer it chose.
AI for Science
Prediction
-
My AI futures forecast — P(doom), P(utopia), and the timeline
~7% doom · ~18% utopia · ~69% friction · ~6% severe. Joint Monte Carlo (ai_futures_sim), one capability spine, 61 events, cruxes before numbers.
-
AGI timeline forecasts all converge on 2027–2028!?
Six quantitative tracks and the AI 2027 scenario land on the late 2020s — plus why forecasts sound urgent while people around you stay calm.
AI control
utopia
interpretability
-
A map of mechanistic interpretability: observe, intervene, validate
SAE, steering, NLA, ACDC, and linear probes feel intuitive because they are variants of the same measurement pipeline — not because the field ran out of ideas. Here is how the tools fit together, what 2026 SOTA actually looks like, and where the hard problems moved.
-
Jailbreak an Open-Source LLM in One Hour — then we tried to make tampering brick the model
Refusal-direction ablation plus an evil system persona pushes Qwen2.5-7B to 92% harmful compliance on HarmBench. Guardrail defenses don't fix it. We trained LoRA adapters that entangle safety with capability — one worked, one only half did.
Psychology
-
Beginner Notes on Meditation: Happier, and a Bit Sharper
A habit that can pay off for a lifetime
-
What Actually Makes Humans Happy
Harvard tracked 2,500 people for 87 years. Neuroscience says dopamine is about wanting, not pleasure. Happy people take more risks. And Nick Bostrom argues that a solved world might be the hardest place to find meaning.
Society
-
How the hippies came alive — and how they died
From Burning Man, San Francisco streets, and Steve Jobs: the hippie movement's Eastern turn, wet vs dry values, what survived and what failed — and why we still live between hard-boiled wonderland and the end of the world.
-
The wealth gap is at a 35-year high. So why does everyone keep buying?
Fed data on the top 1%, Bernays and the invention of desire, hedonic treadmills, and why Nordic countries prove inequality is mostly a policy choice — not human nature.
-
When perfection is impossible: some structural problems in society and AI safety
People complain about government and markets; AI safety talk often jumps to arms races and game theory. Economists and others wrote down more concrete problems and patterns: how opinions get merged, how metrics get gamed, why rules always have gaps.
How AI actually affects work today
-
AI and Writing
Why I started this blog, and how I actually use AI when writing
-
How AI actually works in healthcare
A look at what AI is doing across healthcare in 2026 — where it's delivering real results, where the evidence is more complicated than the pitch, and why the stakes make both sides matter.
-
AI Made You Score Higher. You Didn't Learn Anything.
Students with unrestricted ChatGPT improved practice scores by 48%, but performed 17% worse than controls after AI was removed. A version that gave hints instead of answers eliminated the problem entirely. The key to learning isn't the answer — it's the struggle.
-
AI Writes Half Our Code. We're Working Harder Than Ever.
AI generates 46% of code in enabled files. Controlled studies show experienced developers are 19% slower with it. The software industry is living through the Jevons Paradox in real time.