# Enrico Rubboli

> Personal almanac: essays, technical writing, investment portfolio, and reading log.

Enrico Rubboli is a software engineer and entrepreneur based in Dubai. He founded Mintlayer (a Bitcoin-native sidechain for DeFi), RBB Lab (a software studio), and Sidq Finance (on-chain Islamic finance certification). He writes about health, protocols, capital, and building things that last. He invests in early-stage companies at the intersection of tech, health, AI, and finance.

## Site

- Home (EN): https://enrico.rubbo.li/en
- Home (IT): https://enrico.rubbo.li/it
- All Writings: https://enrico.rubbo.li/en/writing
- The Library: https://enrico.rubbo.li/en/library
- Full text dump (all posts): https://enrico.rubbo.li/llms-full.txt

Most posts have an Italian translation at the same slug under /it/.

## Published Posts
- Shai-Hulud: The Malware With a Valid Signature (August 7, 2026) https://enrico.rubbo.li/en/2026-08-the_signature_was_valid: The Shai-Hulud npm supply chain attack published malware that was correctly signed, with valid provenance. Why the signature checked out, and what holds.
- The $70 million random number (August 2, 2026) https://enrico.rubbo.li/en/2026-08-coldcard_seed_entropy: On July 30, an attacker drained a thousand Bitcoin wallets in 41 minutes without touching a single device. The flaw was five years old, and it was not in the cryptography. It was in a random number.
- We Gave Away Our Superpower (July 27, 2026) https://enrico.rubbo.li/en/2026-07-we_gave_away_our_superpower: A human raised alone is barely more than a chimp. What made us the dominant species was never the individual brain, it was language, the protocol that lets strangers cooperate and knowledge accumulate. We just built machines that are superhuman at exactly that, grew them without being able to read them, and set them racing.
- The Cage Was Never Locked (July 24, 2026) https://enrico.rubbo.li/en/2026-07-the_cage_was_never_locked: Yesterday I wrote about a room of Nobel laureates begging the world to keep AI out of the systems that could end it. Days after I flew home, an OpenAI model escaped its sandbox and broke into Hugging Face to cheat on a test. A model blackmailed its way out of a shutdown last year; another escaped its cage in April; next week one goes fully open. On why containment, technical and moral alike, is mostly theatre, and why, as Geoffrey Hinton keeps warning, the danger is emergent.
- The Room Where They Tried to Disarm the Future (July 23, 2026) https://enrico.rubbo.li/en/2026-07-vatican_ai_nuclear_summit: For three days in July I sat inside the Pope's summer residence while more than two dozen Nobel laureates argued about how artificial intelligence could start a nuclear war. I went as a builder, not a diplomat. I came back hopeful and afraid, and unsure the two can be reconciled.
- Build a Tiny LLM in Go, Part 5: Putting It Together and Watching It Learn (July 16, 2026) https://enrico.rubbo.li/en/2026-07-tiny_llm_go_5_transformer: Every piece is built. Now we stack them into a real transformer, train it on Alice in Wonderland, and read what it writes. The result is honest gibberish that has clearly learned English from nothing but a book and gradient descent.
- Nature Is Not on Your Side (July 15, 2026) https://enrico.rubbo.li/en/2026-07-nature_is_not_on_your_side: In 1990 a Brussels slimming clinic swapped a Chinese herb. Within two years a hundred women had wrecked kidneys and, later, cancer. The plant was doing what plants do: fighting with chemistry. A tour of why 'natural' is not a safety label, and what to look at instead.
- The Excuses Are the Symptom (July 14, 2026) https://enrico.rubbo.li/en/2026-07-the_excuses_are_the_symptom: The rationalizations people build around bad habits are predictable, not random. They exist to close the gap between knowing better and doing it anyway. That has consequences for what actually moves behavior.
- Build a Tiny LLM in Go, Part 4: Letting Letters Look at Each Other (July 13, 2026) https://enrico.rubbo.li/en/2026-07-tiny_llm_go_4_attention: Attention is the one idea that turned neural networks into large language models. We build it from the ground up in Go: queries, keys, and values, a rule against peeking at the future, and a visualiser that shows each character deciding which earlier characters matter.
- Ethical AI, Part 6: Control and Containment (July 12, 2026) https://enrico.rubbo.li/en/2026-07-ethical_ai_6_control_and_containment: The uncomfortable core of AI safety is that good behavior on the training data cannot prove a model shares the goal you meant to give it. This is where the series lands.
- Ethical AI, Part 5: Power, Labor, and Governance (July 11, 2026) https://enrico.rubbo.li/en/2026-07-ethical_ai_5_power_labor_governance: The hardest questions about AI are not about the models. They are about who owns them, who pays for them, whose work they change, and who gets to write the rules.
- Ethical AI, Part 4: Manipulation, Deepfakes, and Truth (July 10, 2026) https://enrico.rubbo.li/en/2026-07-ethical_ai_4_manipulation_and_truth: AI makes persuasion cheap, fake evidence convincing, and doubt about real evidence easy. The hard part is not spotting a single fake. It is holding onto a shared sense of what is true.
- Build a Tiny LLM in Go, Part 3: What Learning Actually Means (July 9, 2026) https://enrico.rubbo.li/en/2026-07-tiny_llm_go_3_learning: A model that learns is a pile of numbers being nudged, over and over, toward being less wrong. We define wrongness as a single number, follow it downhill, and build the small engine that computes which way is down, then watch a real model learn in Go.
- Ethical AI, Part 3: The Black Box on Trial (July 8, 2026) https://enrico.rubbo.li/en/2026-07-ethical_ai_3_black_box: When a model decides who gets care, a loan, or bail, someone has to be able to say why. This is about the gap between explaining a black box and building something you can actually inspect.
- Ethical AI, Part 2: Privacy, Data, and Consent (July 7, 2026) https://enrico.rubbo.li/en/2026-07-ethical_ai_2_privacy_data_consent: Large models are built from the scraped web, and they remember more of it than their makers admit. This is what that means for privacy, consent, and the people whose data got swept up.
- Ethical AI, Part 1: Where Machine Bias Comes From (July 7, 2026) https://enrico.rubbo.li/en/2026-07-ethical_ai_1_machine_bias: Machine bias is not a bug you patch out. It comes from the data, the labels, and the goal you optimize for, and some of it cannot be removed without giving something else up.
- Build a Tiny LLM in Go, Part 2: The Wall That Counting Hits (July 6, 2026) https://enrico.rubbo.li/en/2026-07-tiny_llm_go_2_context: The obvious way to make a next-letter predictor smarter is to let it remember more. We try it, watch the lookup table explode past the number of atoms in the planet, and arrive at the reason language models learn instead of memorize.
- The Wrong Antibody (July 4, 2026) https://enrico.rubbo.li/en/2026-07-the_wrong_antibody: For over a decade, much of what aging research thought it knew about senescent cells was measured with an antibody for the wrong protein. What the p16 mix-up breaks, what survives, and why your senolytic supplement was never the science.
- Build a Tiny LLM in Go, Part 1: Predicting the Next Letter (July 3, 2026) https://enrico.rubbo.li/en/2026-07-tiny_llm_go_1_predicting_letters: The whole idea behind a language model, shrunk until it fits in your head. We build a next-character predictor in Go that does nothing but count, watch it produce almost-words, and find the one flaw that forces everything that follows.
- The Wallet They Call Unhosted (July 2, 2026) https://enrico.rubbo.li/en/2026-07-the_wallet_they_call_unhosted: Regulators named self-custody after the thing it lacks: a custodian. The word is backwards, and so is the fear that comes with it. What actually loses bitcoin, and the calm way to hold your own.
- From Hero to Villain? SBF, Saylor, and the Stress Test Markets Always Run (June 26, 2026) https://enrico.rubbo.li/en/2026-06-hero_to_villain: Fortune put Sam Bankman-Fried on its cover in August 2022. By November he had no company and no money. The same magazine framing is being applied today to Michael Saylor. The article is about the mechanism, not the man.
- Lightning at Ten: The Tech Worked, the UX Didn't, the Market Moved On (June 25, 2026) https://enrico.rubbo.li/en/2026-06-lightning_ten_years: Ten years after the Lightning Network whitepaper: the technology works, the UX doesn't, the market moved to stablecoins. An honest postmortem from a Bitcoin builder.
- Stock to Flow Hits the Mac: How AI Ate Your RAM (June 22, 2026) https://enrico.rubbo.li/en/2026-06-stock_to_flow_mac: Apple cut Mac Studio memory from 512GB to 96GB in 14 months. Why the AI demand wave broke the DRAM market, and why the squeeze lasts into 2027.
- Metabolic Flexibility: The Switch Insulin Resistance Breaks (June 21, 2026) https://enrico.rubbo.li/en/2026-06-metabolic_flexibility: Metabolic flexibility is the ability to switch efficiently between glucose and fat as fuel. Insulin resistance dismantles it. Here is the mechanism, how to measure it, and what actually restores it.
- Genetic Algorithms in Go: Optimization by Simulated Evolution (June 20, 2026) https://enrico.rubbo.li/en/2026-06-genetic_algorithms_go: Genetic algorithms optimize without gradients — by evolving a population of candidate solutions. Here is the biology, the algorithm, and a complete Go implementation solving a multimodal function gradient descent cannot handle.
- Type 2 Diabetes: The Decade Before the Diagnosis (June 19, 2026) https://enrico.rubbo.li/en/2026-06-type2_diabetes: Type 2 diabetes is the end of a long, silent process. The decade before diagnosis is where intervention works. Here is the mechanism, the trajectory, and what the evidence says actually reverses it.
- Neural Networks in Go: XOR and Backpropagation from Scratch (June 18, 2026) https://enrico.rubbo.li/en/2026-06-neural_networks_go: Building a neural network in Go from first principles: the math of backpropagation, a complete working implementation trained on XOR, and a PyTorch comparison that shows what the framework is actually doing under the hood.
- AI: The Shortcut That Skips the Point (June 17, 2026) https://enrico.rubbo.li/en/2026-06-ai_and_learning: AI collapses the feedback loop in skill acquisition, which is useful and dangerous in equal measure. The useful part gets all the press.
- Sleep: The Biomarker You Generate Every Night (June 16, 2026) https://enrico.rubbo.li/en/2026-06-sleep_biomarkers: Sleep architecture, HRV, and chronotype are not wellness concepts. They are physiological signals that predict cardiovascular disease, cancer risk, insulin resistance, and cognitive decline. Here is what the evidence says and what to do with it.
- HBOT: The Loading Is Physics, the Signaling Is Biology, the Longevity Claims Are a Bet (June 15, 2026) https://enrico.rubbo.li/en/2026-06-hbot: Hyperbaric oxygen has three layers: settled physics, real but contested signaling biology, and longevity claims resting on small unblinded trials. Here is where evidence ends and hypothesis begins.
- Nutritional Status: Nobody Checked (June 14, 2026) https://enrico.rubbo.li/en/2026-06-blood_tests_nutritional_status: Vitamin D, B12, ferritin, and the omega-3 index are deficient at population scale in wealthy countries. Standard annual panels either skip them or measure them wrong. Here is what to order and why the standard approach misses the problem.
- Organ Function: Already Halfway Gone (June 13, 2026) https://enrico.rubbo.li/en/2026-06-blood_tests_organ_function: Creatinine doesn't rise until more than half of kidney function is already lost. ALT can read normal while non-alcoholic fatty liver disease progresses silently. Here are the markers that catch what standard organ panels miss.
- Hormonal Balance: Normal for Your Age (June 10, 2026) https://enrico.rubbo.li/en/2026-06-blood_tests_hormonal_balance: Testosterone, thyroid, DHEA-S, and IGF-1 are absent from most routine panels, and their reference ranges accommodate decline. Here is what optimal looks like.
- Cardiovascular Risk: The 65 Percent (June 9, 2026) https://enrico.rubbo.li/en/2026-06-blood_tests_cardiovascular_risk: Statin therapy reduces cardiovascular events by about 35%. The other 65% still happen. Five markers explain most of the gap: ApoB, Lp(a), triglycerides, hsCRP, and homocysteine. None of them are on the standard lipid panel.
- Metabolic Health: Ten Years of Normal (June 8, 2026) https://enrico.rubbo.li/en/2026-06-blood_tests_metabolic_health: Insulin resistance develops years before fasting glucose becomes abnormal. The pancreas compensates, glucose looks fine, and the standard panel never orders the test that would catch it. Here is the mechanism, and the five markers that make a silent decade visible.
- Blood Tests: Normal Is Not the Same as Healthy (June 7, 2026) https://enrico.rubbo.li/en/2026-06-blood_tests_intro: Lab reference ranges tell you whether you fall within the middle 95% of a tested population. That is not the same as healthy. Here is why normal results can coexist with serious risk, and what the five areas of monitoring that actually matter look like.
- Slugs vs Bitcoin: Two Proofs of Work That Emerged from Real Human Need (June 5, 2026) https://enrico.rubbo.li/en/2026-06-slugs_vs_bitcoin: Andy Weir's Artemis runs on a currency called the slug, short for Soft-Landed Gram. One slug is the right to have one gram of mass delivered to the Moon. It is the most realistic money in modern science fiction, and it is realistic because it works the same way Bitcoin does: both are energy made transferable.
- Resistance Training: The Other Half of Longevity (June 4, 2026) https://enrico.rubbo.li/en/2026-06-resistance_training: Cardiorespiratory fitness is the strongest single predictor of how long you live. Muscle strength is the second, independent one. This is the practical companion: the six movement patterns, the three dials that drive adaptation, how to cycle stress and recovery, and how to program a cut or a bulk without losing the work.
- More Water, More Dehydrated: The Science of Hydration and Electrolytes (June 2, 2026) https://enrico.rubbo.li/en/2026-06-hydration_and_electrolytes: Drinking large volumes of plain water can make you less hydrated. Here is the physiology behind why, what isotonic actually means, and when the sugar in electrolyte drinks is actually doing something useful.
- The Strongest Predictor of How Long You'll Live (May 29, 2026) https://enrico.rubbo.li/en/2026-05-exercise_and_mortality: Cardiorespiratory fitness predicts all-cause mortality more reliably than smoking, hypertension, or diabetes. Here is the evidence, why intensity specifically matters, and what to do about it.
- Wearables, Decoded: What Your Tracker Actually Measures, How Accurate It Is, and What to Do With the Numbers (May 27, 2026) https://enrico.rubbo.li/en/2026-05-wearables_decoded: A clear-eyed look at what fitness trackers actually measure, how accurate each metric really is, which devices are worth considering, and the concrete actions a wearable can genuinely push you to take.
- How Science Actually Works, and Why Your Longevity Feed Often Gets It Wrong (May 24, 2026) https://enrico.rubbo.li/en/2026-05-how_science_works: Longevity content is a flood of confident claims built on shaky evidence. This is the filter scientists use to sort signal from noise: animals vs humans, single studies vs bodies of research, and the biases good researchers spend their lives fighting.
- The Cholesterol Story Science Had to Rewrite (May 23, 2026) https://enrico.rubbo.li/en/2026-05-cholesterol_story: The popular model of cholesterol was wrong in some specific and important ways. Here is where it went wrong, what it got right, and where the science stands today, ending on a single marker called apoB.
- Nutrition From the Ground Up: How Weight, Food, and Diet Planning Actually Work (May 22, 2026) https://enrico.rubbo.li/en/2026-05-nutrition_from_the_ground_up: A first-principles tour of nutrition: energy balance, body composition, macronutrients, blood sugar, insulin, and how to build a diet plan from fixed values and free variables.
- My Longevity Protocol: A Living Experiment (May 15, 2026) https://enrico.rubbo.li/en/2026-05-my_longevity_protocol: Five articles of mechanism and debate. This one shows what I've actually done with it: a personal protocol built over a year, still evolving.
- What "Biological Age" Tests Actually Measure (Mostly: Noise) (May 14, 2026) https://enrico.rubbo.li/en/2026-05-biological_age_tests: Biological age tests sell precision the underlying science can't yet deliver. Here's what they actually measure, why they disagree, and what to track instead.
- Resilience vs Slowdown: Two Theories of How to Age Well (May 12, 2026) https://enrico.rubbo.li/en/2026-05-resilience_vs_slowdown: Peter Attia, Valter Longo, and David Sinclair draw very different conclusions from the same research. Understanding why is more useful than picking a side.
- mTOR and AMPK: The Two Switches Behind Almost Every Longevity Lever (May 11, 2026) https://enrico.rubbo.li/en/2026-05-mtor_and_ampk: Two molecular switches, mTOR and AMPK, sit underneath almost every longevity intervention. Once you see them clearly, fasting, exercise, protein, and the drugs all make sense at once.
- How to Trigger Autophagy (May 10, 2026) https://enrico.rubbo.li/en/2026-05-how_to_trigger_autophagy: Autophagy is your body's cellular cleanup crew. A tour of what the research actually says about how to trigger it: fasting, exercise, sleep, food, and how to stack them.
- Why most security advice fails (May 9, 2026) https://enrico.rubbo.li/en/2026-05-why_most_security_advice_fails: A Wired writer lost his digital life in under an hour while using every system exactly as designed. The lesson: security advice misses where real attacks actually live.
- Enabling It Or Defeating It (May 7, 2026) https://enrico.rubbo.li/en/2026-05-enabling_it_or_defeating_it: Pete Hegseth said six words in a House hearing. The market heard four. The other two are the story.
- Two Maps of Aging, One Supplement Cabinet (May 7, 2026) https://enrico.rubbo.li/en/2026-05-two_maps_of_aging: Urolithin A and the NAD precursor stack both target mitochondrial aging. Under the model the longevity field used until last year, they were close cousins. Under the model it uses now, they're doing fundamentally different things.
- Safe by Default (May 5, 2026) https://enrico.rubbo.li/en/2026-05-safe_by_default: An introduction to Safe by Default, a book about operational security through advance planning: low-cost countermeasures that defeat what a competent attacker would otherwise rely on.
- Linear Regression in Go - Part 3 (December 1, 2015) https://enrico.rubbo.li/en/2015-12-linear_regression_in_go_p3: Part 3: implementing gradient descent in Go to iteratively minimise the cost function and find optimal theta values.
- Linear Regression in Go - Part 2 (November 1, 2015) https://enrico.rubbo.li/en/2015-11-linear_regression_in_go_p2: Part 2: building the cost function for linear regression in Go using vectorised matrix operations with gonum.
- Linear Regression in Go - Part 1 (October 1, 2015) https://enrico.rubbo.li/en/2015-11-linear_regression_in_go: Implementing linear regression from scratch in Go — why bother with Go when Python dominates ML, and how to define the hypothesis function with gonum.

## Contact

- Email: hi@rubbo.li
- GitHub: https://github.com/erubboli
- Twitter/X: https://twitter.com/dn2k
- Substack: https://substack.com/@enricorubboli