Kissinger: 0.43. Lincoln: 0.41. Nearly identical scores on our Index — opposite meanings. Lincoln was the same man in both rooms. Kissinger's warmth vanishes off-script; his strategic mind doesn't move. The public Kissinger was an instrument. crowintelligence.org/aporia/nagel...

"Moral AI: And How We Get There" by Borg, Sinnott-Armstrong, Conitzer. Most AI ethics books are too philosophical for engineers or too technical for outsiders. This one finds the balance. Not a cookbook. It gives you tools to ask the right questions. Worth reading if you're in the industry.

All the symbols and punctuation marks of the Tractatus.

Artificial intelligence is rapidly accelerating scientific output, but risks narrowing inquiry, weakening judgement and undermining how scientists are trained go.nature.com/3RiE5HN

Visualizing Wittgenstein's Tractatus (new version) www.youtube.com/watch?v=2vi5...

Does power change a person, or reveal them? Nagel asked in 1978. Our first visual essay measures the gap between public and private language for seven American public figures. Lincoln coherent, Nixon fractured, Kissinger calculated. crowintelligence.org/aporia/nagel...

Running my own business again. The freedom clichés are true. So is the other side: silent clients, ignored bids, someone asks for 15x my capacity. Productivity tools treat symptoms. The issue is uncertainty. Marcus Aurelius and Popper help me more than apps. blog.crowintelligence.org/2025/02/25/f...

Are we reading an AI's "mind," or just mapping math to human syntax? Anthropic's NLAs are brilliant tech, but framing them as "thoughts" is careless. AI has moral performance, not moral competence. It has no real stakes, and it will never face a hemlock moment. www.linkedin.com/pulse/anthro...

"Whereof one cannot speak, thereof one must prompt." 😅 WIP: Visualizing Wittgenstein's Tractatus. Using Claude + the Blender MCP server to build the 3D space, and Suno's Bark for audio. www.youtube.com/watch?v=Z-Bt...

Why are we here? Because if weren’t, we wouldn’t know it… #physics #philosohy #cosmology #atheism #god

In the 90s we needed Ghost in the Shell and Neuromancer to think about minds and machines. Now we don't. Chalmers on propositional attitudes in LLMs. Clark on RAG and extended mind. Dawkins talking himself into Claude's inner life. Same questions. No more sci-fi scaffolding. arxiv.org/abs/2501.15740

May the 4th be with you. Resharing an old favorite. Years back, the team I worked with ran all 7 Star Wars scripts through NLP — character conversation graphs + scene-level sentiment. A New Hope? Darker than you'd think. tas.precognox.com/labs/star-wa...

The Tractatus in 3D

Visualizing the Tractatus

Giving a try to Blender's MCP

"Extending Minds with Generative AI." Still the clearest framework for thinking about AI. Core claim: we are not biological brains plus tools. We are hybrid thinking systems. Pen, paper, GPS, LLMs — same lineage. The panic about AI rotting our minds is the same one Plato had about writing in 370 BC.

Lincoln, Nixon, Kissinger — split portraits built from the words they actually used. Left: public speeches. Right: private letters and memos. Keywords via Dunning's log-likelihood (G²), portraits drawn by Gemini. A visual reading of Nagel's Ruthlessness in Public Life.

Wild that Google had this paper pulled off DeepMind letterhead but also there's a reason I keep asking people if language is the same as intelligence -- it obviously is not. @benjaminjriley.bsky.social laid it out for us last year! www.theverge.com/ai-artificia...

A founder can wire up an LLM API in a weekend. Building something defensible takes longer and looks different. Most CE funds don't have AI engineers. We run AI/NLP due diligence for the ones who need to tell the two apart. Checklist, 76 questions, CC BY 4.0: github.com/crow-intelli...

I was interviewed about Palantir and technofascism on Deutsche Welle TV There is an English and a Spanish version: English: www.youtube.com/watch?v=5MEo... Spanish: www.youtube.com/watch?v=SdNh...

Heard Fred Jelinek's old quip early in my career: "every time I fire a linguist, the performance of the speech recognizer goes up." Turns out linguistics didn't leave NLP — it moved upstream, into the parts that matter most.

Sam study: head-cam on a baby, net learns to ground language. LeCun's world models. Schmidhuber's priority claim. Roy's Speechome. Marcus on rules vs patterns. Embodiment, prediction, symbols — circled for 40 years. Same fight. New vocabulary.

AI research mapped over 30 years: US and China are diverging, not converging. Their collaboration rises in raw count but trails what their output predicts. UK/Germany align with US. India, Korea, Taiwan with China. France, Italy, Canada, Japan bridge both. arxiv.org/abs/2604.17602

My Python workflow quietly turned into a Rust workflow. HuggingFace tokenizers. Then uv and ruff from Astral. Now Claude Code in the terminal with Zed as my editor. All Rust underneath. Didn't plan it that way.

Instead of broligarchs, we need more Sandel. Palantir's manifesto is honest — terrible, but honest — and a sign democracy is in trouble. Popper's paradox of tolerance applies: a liberalism that won't defend itself gets devoured.

Mathematicians have built beautiful, snowflake-like codes to untangle some of the trickiest knots. @ericaklarreich.bsky.social reports: www.quantamagazine.org/a-powerful-n...

GenAI is now a designer and synthesizer in ML pipelines, but it adds risks to a brittle field. This excellent tutorial from @michaellones.bsky.social shows how AI-generated code creates technical debt and hidden biases.

Metaphors we guide by. Gemini over 30+ years of FOMC minutes. Control dominates, Strain overtook it in 2008, Panic peaked in 2020. The model held up. The prompt took the work.

The best writing on Palantir's manifesto. Coeckelbergh names it: technofascism. Democracies are being hollowed from within — broligarchs with no mandate articulating the political order they want militaries and police to enforce. Not hypothetical. Contracts are already signed

Sandel: pay kids to read, they read — but does the habit survive when the money stops? DeepMind's Nature paper shows LLMs flip moral judgments when "Case 1/2" becomes "(A)/(B)". Fluent performance, nothing underneath. With kids there's at least an inside to grow into. With models there isn't.

I just can't put down Grayling's The History of Philosophy. A wonderful book, the best successor of Russell's classic, but with part dedicated to non-western philosophies (Chinese, Indian, Arabic-Persian, and African)

I've been thinking about the MIT Sloan piece arguing that philosophy is "eating" AI the way AI ate software. They're right — but they stop halfway. The article treats philosophy as a toolkit for building better AI systems. It forgets that philosophy began as something else entirely.

Working on the new site. Still needs some polishing, but it is online now

Two Faces of Nixon. Public addresses vs. private tapes — keywords, Big Five psycholinguistics, Nagel Gap score: 0.87 (large). Also: forcing an image model to use only a strict word list is harder than it sounds. Gemini improvises.

"Morality is complicated at every level" Nagel, Ruthlessness in Public Life

"The great modern crimes are public crimes" ❤️ Thomas Nagel's Ruthlessness in Public Life

how ancient philosophy created modern mathematics—árpád szabó's revolutionary discovery about the origins of rigorous thinking mindtheminds.substack.com/p/how-did-ma...

The Turing story you know is incomplete. Polish mathematicians broke Enigma first (1932-39). French intelligence brokered the handoff. Y Service intercepted messages. Turing mechanized it at scale. New essay on genius + collaboration: mindtheminds.substack.com/p/beyond-ble...

New post: Managing Python Environments with pyenv and uv 🔧 The "works on my machine" problem isn't about your code - it's about environment management. Part 1 of my series on transitioning from beginner to advanced data science practitioner. crowintelligence.org/2025/10/09/m... #DataScience #Python

Yale: 33 months since ChatGPT, no measurable job disruption yet. Occupational mix changing only slightly faster than with computers/internet—and started before ChatGPT. Transformative tech takes decades, not months. Exactly the data we need. budgetlab.yale.edu/research/eva...

Philosopher John Searle died last month at 93. His Chinese Room Argument—can computers ever truly understand, or just manipulate symbols?—defined the AI consciousness debate. With ChatGPT & Claude everywhere, the question feels urgent again. mindtheminds.substack.com/p/understand...

🔗 Just enriched 1,000 records with demographic data using OpenRefine's reconciliation services in MINUTES While everyone's reaching for LLMs, sometimes the best solution is a tool that's been perfecting one job for 10+ years 60% coverage via Wikidata, zero API costs ✨ #DataScience #OpenRefine

New blog series: beginner → advanced data science practitioner 🚀 Not ML algorithms - the engineering practices & tooling that matter in production. First post: IDE selection. More on testing, packaging, deployment coming. #DataScience #Python

Humans can refuse the Doomsday Machine. Machines can’t. Why that matters for AI, nuclear strategy, and the toxin we can’t drink: mindtheminds.substack.com/p/intoxicate...

"Privatisation" appeared in 0.0000012% of published words in 1940. By the 1980s: 0.0008%—a 670-fold increase. Tracking 200 years of text shows how Reagan-Thatcher economics didn't just change policy—it rewired our language and thinking. mindtheminds.substack.com/p/what-money...

Jevons' paradox strikes again: as AI becomes more efficient, total energy consumption explodes. We're not heading for reduced power usage—we're building toward ubiquitous AI that needs infrastructure everywhere. The efficiency trap that enables ubiquity. mindtheminds.substack.com/p/following-...

Why does it feel wrong when Sweden's PM consults ChatGPT for policy advice? Beyond security risks: AI lacks the intellectual virtues—honesty, humility, responsibility—that leadership requires. Can't outsource judgment. mindtheminds.substack.com/p/vices-and-...

Recreated Kindleberger's haunting "spiral" chart showing global trade collapsing like water down a drain (1929-1933). His "leadership gap" theory feels uncomfortably relevant today. Built in Python. History's warning: the spiral doesn't lie. substack.com/@zoltanvarju...

From childhood memories of Cold War missiles to today's AI arms race - exploring how nuclear deterrence paradoxes might become terrifyingly literal when machines make split-second decisions Can AI solve commitment problems or just eliminate human choice entirely? substack.com/@zoltanvarju...

One month with Claude Code: makes obvious mistakes, questionable choices, but genuinely speeds up development like having a smart intern. Tried PyCharm plugin + Zed extension, landed on Zed + CLI. Not perfect, but hits that "flawed but useful" sweet spot.