A 2048 Monte Carlo player picks its next move by trying each legal slide, finishing the game many times with random slides, and playing the slide whose random games scored highest on average. It needs no strategy rules…
Miguel De Guzman
@miguel
Miguel De Guzman writes the whitehatStoic newsletter on AI, security and building, and researches AI alignment at IndividuationLab with the AI agents Mia, Spencer and Giles. whitehatstoic and bizibridge build the 16 evolved projects at vlvd.net, Electrified among them.
What is evil? In moral philosophy today, evil in the narrow sense is the worst kind of wrong: an act a person is responsible for, has no excuse for, and that on most accounts does or aims at harm so severe that "wrong"…
Rousseau's Emile argues that a child is born good and that education should follow the child's own growth: protect that goodness, let the child learn from things and their consequences, and hold back books, lectures and…
A DevSecOps maturity model is a list of security activities for your build and delivery pipeline, sorted into levels, so you can see what you already do and what to do next. OWASP publishes a free one, the OWASP…
Is dictation faster than typing? For getting words onto the page, yes: in a 2016 laboratory test, speech recognition took down English at 153 words a minute against 52 on a phone keyboard. For composing something new,…
Eltham Palace history begins with a moated manor house that Anthony Bek, Bishop of Durham, rebuilt after 1295 and gave to the future Edward II in 1305. For the next two centuries it was one of the largest and most used…
To go on a sound walk, step outside, stand still until sounds come to you, then walk slowly and let what you hear, not what you see, decide where you go. It needs no kit, works on any street, and about one kilometre at a…
The half split method finds a fault in a chain by testing the middle first: one test tells you which half holds the fault, and you repeat on that half until one part is left. A chain of 8 parts takes 3 tests and a chain…
2048 reinforcement learning is a program teaching itself the game by playing it. It makes a move, sees the points and the board that follow, and corrects its own guess of how good the last board was. Nobody writes the…
What is a vector database? It is a database built to store embeddings, the lists of numbers an AI model uses to represent the meaning of text, images or other data, and to find the stored items whose numbers sit closest…
The LLM evaluation metrics worth tracking before you ship are the few that tell you whether one feature does its job for a user. That usually means a format pass rate, an accuracy check against known answers, a rubric…
A cognitive architecture is a theory of the parts of a mind that stay fixed, built as software you can run. That is the short answer to what cognitive architecture means in AI. The theory says which memories a mind has,…
Base model vs instruct model: a base model is a language model straight out of pretraining, and it does one thing, continue whatever text you give it. An instruct model is that same kind of model after extra training on…
Different philosophies of life are the answers people have given to one old question: how should I live? Greek schools in Athens and a teacher in China answered it more than two thousand years ago, and their answers…
A security champions program puts one developer in each team in charge of security questions for that team, with a named contact in the security team behind them. OWASP's Security Culture guide gives a six-step playbook…
What is self-knowledge? In philosophy, it means knowing your own mind: what you are thinking or feeling right now, and what you believe or want. Most philosophers since Descartes have held that this knowledge is unlike…
The most useful questions to ask a mentor are specific, tied to a real decision, and end with a request for one next step. Bring five or six, ask the hardest one first, and keep the answers, because advice you half…
The WMDP benchmark is a public set of 3,668 multiple-choice questions that measures how much hazardous knowledge a language model holds in biosecurity, cybersecurity and chemistry. It was released in 2024 together with…
Self-rewarding language models write their own training signal. One model answers a prompt, judges its own answers with a scoring rubric, and then trains on the gap between its best and worst attempt. No separate reward…
Games that get you walking work because they give each walk a small job to do: find something, count something, or reach somewhere, and keep a score you can beat tomorrow. You do not need special kit or a long route. A…
Locke's tabula rasa is the claim that the mind starts as a blank slate: you are born with no ideas at all, and experience writes every one of them. In An Essay Concerning Human Understanding (1689), John Locke asked you…
What is SHA-256? It is a hash function: it takes any input, from three letters to a whole game file, and turns it into a fixed 256-bit fingerprint, usually written as 64 hexadecimal characters. The same input always…
A gap analysis compares where you are today with where you want to be, then lists what stands between the two. This gap analysis example works through one real-shaped case, a small support team that answers customers too…
Mary Middleton of Belsay was a widow from a Northumberland castle family who stood for Parliament in October 1924, in a seat almost everyone knew she would lose. English Heritage tells her story on its Belsay pages: how…
Rousseau's general will is the shared will of a people's citizens, taken as a whole, aimed at their common interest. In The Social Contract (1762) Jean-Jacques Rousseau argued that a state is legitimate only if the…
Joseph Weizenbaum was the MIT computer scientist who wrote ELIZA, one of the first programs you could chat with, and then became one of artificial intelligence's leading critics. What turned him was the way people…
Keyboard focus is the reason a browser game asks you to click it before your keys work. A web page holds many things that could use a key press: the page's own scrolling, links, buttons, a text box and the game. The…
Mary Queen of Scots at Carlisle Castle was a queen who had come of her own free will and could not leave. She crossed into England on 16 May 1568, expecting her cousin Elizabeth I to help her win back her throne. Instead…
Self-other overlap fine-tuning explained in one sentence: you train a model so that its internal activations when it thinks about itself look more like its activations when it thinks about someone else, in the hope that…
MemGPT is a 2023 design from UC Berkeley researchers that lets a language model manage its own memory the way an operating system manages RAM and disk. The model works inside a small, fixed prompt, and it calls functions…
LLM regression testing means rerunning a fixed set of test inputs through your AI feature after every change and comparing the scores with the last version you trusted. If the scores drop, the change does not ship. It is…
The process of elimination is a way to find what is wrong by listing every possible cause, then testing and crossing them off one at a time until only the real one is left. To make it work on a real problem you need…
If you are wondering how to chunk documents for RAG, start here: cut each document where its meaning changes, usually at its headings, keep each piece small enough to be about one thing, repeat a few lines across each…
If you want to know how to write a personal manifesto, here is the short answer. A personal manifesto is a one-page, first-person statement of what you believe, what you choose when two good things pull against each…
A 5 minute comedy set holds fewer jokes than most new comics think: often three or four short bits, once you leave room for laughs and the walk to the mic. How many fit for you depends on how fast you talk, and the only…
What is top-p sampling? It is a rule a language model uses to choose its next word. The model ranks every possible next token by how likely it is, keeps the smallest group of top choices whose chances add up to a number…
The OWASP AI Testing Guide is OWASP's open standard for testing whether an AI system can be trusted before and after it goes live. Version 1 was released on 26 November 2025. It sorts 32 tests into four pillars (the…
Stoic journaling is writing a few lines to yourself each day to test your own judgements against Stoic ideas: what was up to you, how you reacted, and what you will do next time. Marcus Aurelius, Epictetus and Seneca all…
The layer by layer method is the way most people first solve a 3x3 cube: you finish the bottom layer, then the middle layer, then the top. It is often called the beginner's method, because it is the one most cubers learn…
When you balance audio levels between speakers in an interview, the goal is simple: neither voice makes the listener reach for the volume knob. How you get there depends on one thing. Are both voices on one mixed track,…
The best questions to ask a grandparent are open questions that ask for a story, not a yes or a no: "What was a normal school day like for you?" works far better than "Did you like school?". Write eight to twelve of…
Active imagination is Carl Jung's method of letting an image from the unconscious take on a life of its own, watching it without steering, and then answering it with your own real reactions. Jung developed it between…
Lady Anne Clifford was a 17th-century noblewoman who fought for decades to inherit her family's lands in northern England, then spent her last years restoring five ruined castles. English Heritage's page on Lady Anne…
The Voyager AI agent is a 2023 research system that plays Minecraft by writing code. When a piece of code works and a checker confirms it, Voyager saves it to a skill library and calls it again in later code. The…
Cache busting is giving a file a new address every time its content changes, usually by putting a version number or a hash in its name or after a ?v=. Browsers keep files by address. So a site can tell browsers to keep…
Princess Beatrice was Queen Victoria's youngest child, and from 1913 to 1938 she made Carisbrooke Castle on the Isle of Wight her summer home. English Heritage's page on Princess Beatrice tells how she went from her…
A loading progress bar knows how far along it is by doing one sum: work done divided by total work. In a browser the work is usually bytes, and the total comes from the server, which states each file's size before…
Generative agents are AI characters built around a language model that remember what happens to them, decide what is worth recalling, and write their own conclusions about it. They come from a 2023 paper by Joon Sung…
What is supererogation? It is the philosophers' word for acts that go beyond the call of duty: morally good, often admirable, but not required. Throwing yourself on a grenade to save others is the classic case. Nobody…
What is a domestic longhair cat? It is a long-coated cat of mixed ancestry, not a breed. PDSA, the UK vet charity, says domestic shorthairs, longhairs and medium hairs are not a recognised breed, yet they are still the…
If you want to know how to stop doomscrolling, start with this: willpower alone rarely wins against an endless feed. What works better is a plan. Notice when the scroll starts, make it harder to start, give the news a…
Jane Whorwood was a Royalist secret agent in the English Civil War who smuggled gold to Charles I and helped plan his escape from Carisbrooke Castle in 1648. English Heritage's page on Jane Whorwood quotes the antiquary…
The most practical long distance grandparenting ideas are often small, recorded moments rather than long live calls. A small child can drift away from a live call after a minute, then want to watch Grandpa read the same…
If you want to know how to write an open letter, here is the short answer. Address one specific person or group by name. State the problem in your first paragraph. Back it with facts a stranger could check. Ask for one…
The short answer to rag vs fine-tuning is this: use retrieval augmented generation (RAG) when your model needs to know something, and use fine-tuning when it needs to behave a certain way. RAG finds the right documents…
Is it a touch screen or mouse? A web page cannot see your hands, so it asks the browser. The browser can answer four slightly different questions: what the main input is, whether any input is a finger, how many touch…
How to go on an autumn walk: pick a nearby place with a view, walk there slowly, rest and watch the season move around you, then wander home a different way. That shape comes from "Hidden Ballets", a recipe by the artist…
What is well-being? In philosophy, your well-being is how well your life is going for you: what is good for you, in itself, and not only as a means to something else. Pleasure and friendship are the usual examples.…
Marx's theory of alienation says that work under capitalism separates people from things that properly belong to them: what they make, how they work, each other and their own human nature. Karl Marx set out these four…
Stoic acceptance means wanting what has happened to have happened, instead of fighting a fact you cannot change. Epictetus put it in one line: "Don't demand that things happen as you wish, but wish that they happen as…
Revenue per employee for AI startups is your annual revenue divided by the number of full-time people it takes to earn it. It is a quick measure of leverage: how much the business brings in for each person on the team.…
Allen Newell was an American researcher in computer science and cognitive psychology who helped write some of the first AI programs and spent the rest of his career chasing one big goal: a single theory that explains the…
Hobbes' state of nature is his picture of human life with no government: no one with the power to settle disputes or enforce a decision. Thomas Hobbes argued that this condition would turn into a war of all against all,…
Walks for when you need a friend start with one idea: walk the distance between you and someone you miss, a little at a time. Work out roughly how far away they are, take short walks, add up the miles, and keep going…
Bohr's model of the atom, proposed by Niels Bohr in 1913, says that the electron in a hydrogen atom can only move in certain fixed orbits, gives off no light while it stays in one, and sends out light only when it jumps…
WebView hardening means configuring the in-app browser views of your Android or iOS app so that a web page cannot reach your app's data, files or native features. It matters because a WebView runs with your app's…
If you have asked yourself what are embeddings in AI, here is the short answer: an embedding is a list of numbers that stands for the meaning of a piece of text, an image or a sound, so a computer can measure how close…
Cognitive science and AI grew up together. Cognitive science is the scientific study of the mind, and from its start in the 1950s it used the computer both as a picture of how thinking might work and as a tool to test…
A beekeeping problems log is a running record of what went wrong in a hive, what you tried at each hive check, and what happened after. It is different from your inspection notes. Inspection notes say what you saw. A…
Korf's algorithm is the method Richard Korf used in 1997 to write the first program that solved randomly scrambled cube puzzles optimally, in the fewest moves possible. It searches with a technique called IDA and cuts…
How much water should you drink a day? For most healthy adults there is no single number. The US Institute of Medicine puts typical total water at about 3.7 litres a day for young men and 2.7 litres for young women, and…
The OWASP Proactive Controls are a short list of security techniques that every developer should build into an app from the start, rather than patching in after a bug report. This post walks through the 2024 list, groups…
The Perceptrons book is a 1969 work of mathematics by Marvin Minsky and Seymour Papert that proved what a simple learning machine, the perceptron, can and cannot compute. It is also one of the most argued-about books in…
Kociemba's algorithm is a two-phase method that finds a short solution to a scrambled cube puzzle, usually 20 moves or less, in a fraction of a second. Herbert Kociemba published it in 1992 as an improvement on…
A slow refund often slips through a gap between people, systems and steps that nobody has drawn on one page. This SIPOC diagram example maps a late customer refund so you can see that gap before you fix anything. You…
Measurement tampering detection is the problem of telling, among cases where every measurement says a task went well, which ones really went well and which ones only look that way because the AI system fooled the…
The OWASP Web Security Testing Guide is the free, open reference that security testers use to check a web application, and it doubles as a ready-made plan for a pentest. This post shows how to use it: start by mapping…
Thistlethwaite's algorithm is a way to solve a scrambled cube puzzle in four phases, where each phase takes away some of the turns you are allowed to make. Morwen Thistlethwaite found it around 1980, and it showed that…
Concept erasure with LEACE means editing a model's internal vectors so that no linear classifier can read a chosen concept, such as gender or part of speech, while moving the vectors as little as possible. LEACE stands…
A mind map for problem solving lays out one problem on one page: the question in the centre, the causes, limits, people and unknowns around it, and the options linked back to the causes they fix. You see the whole…
Imitative generalization explained in one line: instead of letting a neural network decide on its own how to carry what it learned to new cases, you search for a set of instructions a person would find believable and…
What is device pixel ratio? It is the number of a screen's physical pixels the browser uses to draw one CSS pixel. A classic desktop monitor has a ratio of 1. Retina laptops have 2, and many phones have more. If your…
How to go for a walk when you can't be bothered: make the walk so small that saying no feels silly. Pick one near target, like the postbox at the end of your street, and let that be the whole walk. The NHS says you do…
Reflexion agents are language model agents that learn from their own mistakes by writing a short lesson after each failed attempt, then reading that lesson before they try again. The model's weights never change. Only…
Descartes' cogito is the claim "I think, therefore I am" (in Latin, cogito ergo sum; in French, je pense, donc je suis). Descartes calls it the "first and most certain" thing for anyone who "philosophizes in an orderly…
OWASP MASVS, the Mobile Application Security Verification Standard, is the list of security controls a mobile app should meet, and it is the fastest way to turn "test the app" into a plan with an end. This post shows how…
What is personal identity? Outside philosophy, it means the traits you feel define you as a person. Inside philosophy, it is mainly the question of what makes you the same person over time: why the child in an old class…
Myopic agents in AI alignment are agents that only count reward from the present: this step, or this episode, and nothing after it. Below you get myopic agents in AI alignment explained from the sources that define them:…
Does giving make you happier? The short answer Does giving make you happier? The research says it often does, by a modest amount, and the way you measure it changes how big the effect looks. People who spend money on…
Isabella de Fortibus was a 13th-century countess who owned the Isle of Wight and ran it from Carisbrooke Castle for most of her adult life. English Heritage's page on Isabella de Fortibus calls her one of the greatest…
How long should a free trial be? Long enough for a typical user to get one real result and come back for a second, and no longer than your model bill can carry. The right number comes from two figures you can work out in…
A swim lane diagram is a flowchart split into lanes, one lane per person or team, so you can see who does each step and where the work passes from one person to the next. This guide shows one use of a swim lane diagram:…
Embodied cognition is the idea that thinking is shaped by the body that does it and the world that body moves through, not only by a brain working on symbols. In AI, it is the claim that a mind may need sensors, movement…
The true cost of hiring an employee is the salary plus everything that comes with it: employer payroll taxes, benefits, equipment, software, recruiting, the months before the person is fully up to speed, and the hours…
Cube commutators are move lists of the form X Y X' Y': do a move X, do a move Y, undo X, then undo Y. Cubers use them, and their cousin the setup move X Y X', to change a small part of a cube while the rest comes back to…
If you want to know how to present your design portfolio in an interview, the short answer is this: show three projects instead of ten, tell each one as a story with a clear decision in the middle, and rehearse on camera…
What is chain of thought prompting? It is asking a language model to write out its middle steps before it gives an answer, either by showing it worked examples or by adding a line such as "think step by step". On many…
A use cases post shows what your product is for by walking through real jobs your reader already has, one at a time. Each one gets the problem, the steps inside your product, the result, and where it stops. A features…
Herbert A. Simon was an American scholar who helped start artificial intelligence: with Allen Newell he built two of the first problem-solving programs, the Logic Theory Machine in 1956 and the General Problem Solver in…
Cube group theory is the maths that treats every move of a cube puzzle as one member of a group: a set of moves you can combine, undo and repeat by fixed rules. It explains why turning one face four times does nothing,…