Human approval for AI agents is the check where a person says yes before the agent does something that matters: paying an invoice, deleting data, sending an email to a customer. Most teams add a confirm button and call…
A place to write, and be read.
Practical guides from vlvd.net on AI safety, security testing, study skills, history and games. Read them, follow the writers you like, or write your own.
Latest posts
Fresh off the bench.
Newest first, from everyone. Sign in to follow people and see only their posts.
Hidden Unicode prompt injection hides instructions in characters that a screen does not draw. A support ticket can look like "Please reset my password" to you, while the model reading it also sees a second sentence…
What is clock speed? Also called clock rate, it is how many steps a processor takes each second, set by a clock inside the chip that ticks billions of times a second. It is the GHz number on every spec sheet, and it…
Is overclocking safe? It can be done carefully, but it is never risk free. Overclocking means running a computer chip, such as a CPU or a graphics chip, faster than the clock speed its maker rated it for. It can shorten…
Tank City Reboot has six tank game stages, and each one is built to test a different skill. This guide walks through all six: what each map is made of, which malware tanks it sends, and a plan to clear it. Every map and…
Reward model overoptimization is what happens when you train an AI model too hard against a learned reward model: the reward model's score keeps rising, but the quality you actually wanted levels off and then falls. It…
AI sycophancy is when a language model tells you what you seem to want to hear instead of what is true. It praises your essay more because you said you wrote it. It drops a correct answer the moment you push back. It…
Embedded agency is the problem of how an agent should reason and act when it is part of the world it is trying to change. Most theories of rational choice quietly assume the agent sits outside its world, like a player…
Rubik's cube notation uses six letters and three small marks. The letters name the faces: F front, B back, U up, D down, L left and R right. A plain letter turns that face a quarter turn clockwise, a prime (R') turns it…
Playing 2048 on a 3x3 board sounds like an easier warm-up: fewer squares, fewer tiles to track. In practice it is the opposite. With 9 squares instead of 16, the board fills in a few dozen moves, and a simple count shows…
How does 2048 scoring work? The short answer: every time two tiles merge, you get the value of the new tile as points, and nothing else scores. That one rule explains why your score jumps late in a game, why two players…
How to find new walking routes near you? Change one thing about the walk you already do: go the other way round, take a side street, or aim for a park or towpath on the map. You rarely need a car or a trip to the…
Here is how to photograph a black cat with your phone: put the light behind you, pick a background lighter than the cat, tap its eyes to focus and drag the brightness up a little, with no flash and no zoom. Without…
A tuxedo cat is a black and white cat, usually black on top with white on the chest, the paws and often the face, so it looks dressed for dinner. "Tuxedo" is a nickname for the coat, not a breed. This guide explains what…
Knowing what to say in a letter to your grandchildren is harder than it looks. You love them, and that part is easy. Turning it into a page they will want to read in twenty years is not. Below you get eight ideas for…
Will AI care about humans? People ask it in hope and in fear. Most answers jump straight to whether AI will help us or harm us. This post asks a step earlier: what would it even mean for a machine to care, what today's…
How would we know if AI is smarter than us? It sounds like a yes or no question. It is not. Before anyone can say a machine has passed us, they have to say what "smarter" means and how to measure it, and the experts do…
If you are wondering how to remember historical figures and what they did, the trouble is usually not the names. You know you have heard of Gutenberg, Luther and Rousseau. The trouble is the match: which one did what,…
Locke vs Rousseau is one of the classic pairings in political philosophy. Both imagined people living without government, and both said a legitimate state rests on a kind of agreement, a social contract. But they used…
Why was the printing press important? Gutenberg's printing press changed a world where books were copied by hand or printed from carved wooden blocks, so few people could afford one. Johannes Gutenberg's movable metal…
You studied a chapter on Sunday and by Thursday it feels like you never read it. That is not a sign that something is wrong with your memory. Forgetting is normal, and it follows a pattern. This post explains how to stop…
Finals week in college is different from any test before it. You may have four or five exams in a few days, each one covering a whole term, some of it from chapters you read months ago. This guide covers how to study for…
The exam is in two days and you have barely opened the book. Cramming, studying everything in one long session just before a test, is what most students fall back on at some point. So does cramming work? The honest…
A castle drawbridge was a bridge across a castle's moat or ditch that could be raised, so the way in disappeared when danger came. The short answer to how it worked: it was hinged at one end, and a winch in a room above…
The Two Towers ballista is the tower you pick when armor, shields and helpers are what break your gate. It fires a heavy bolt through a line of attackers and ignores armor. This guide covers what the Ballista does, which…
A springald was a medieval siege engine that shot heavy bolts: a bow built into a strong wooden box, with its two arms set in twisted bundles of hair or sinew and drawn back by a screw. Castles and armies used…
AI agent memory security is about what your agent remembers between conversations, and who that memory can reach. Memory makes an agent feel helpful: it knows your name, your plan and what you asked last week. It also…
MCP server security is about one question: what happens when you let an AI agent use a tool someone else wrote? The Model Context Protocol (MCP) is a standard way to connect an agent to tools, files and services. Adding…
AI generated code security is the work of checking code that a coding assistant wrote before it reaches your users. The assistant is fast and often right. But it does not know your threat model, it can invent package…
Study group problem solving works best when every member brings the exact point where they are stuck and the group records each approach and its outcome. A good study session can crack a question nobody could do alone.…
Write here too
Your own page.
Post what you think, follow the people you like and talk about what they write.