If you want to know how to write a post that explains a technical decision, start with the decision itself, then give the reasons in the order a reader needs them. Open with what you chose and why in two sentences. Then…
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.
Why jailbreaks keep working on aligned language models comes down to one fact: alignment training changes what a model tends to do, and it does not remove what the model can do. A jailbreak is a prompt that finds a path…
New team member onboarding problems are easy to notice, mention in passing and forget until the next person hits the same wall. A simple fix is to keep them in one shared fixes list, where every problem has a line, an…
Dunstanburgh Castle on the Northumberland coast was begun in 1313 by Thomas, Earl of Lancaster, who was then the wealthiest nobleman in England. English Heritage says he built it as much to protect himself as to guard…
Can giving an AI agent an identity make it safer? Possibly, but nobody has shown it yet. A defined identity may help an agent keep its boundaries because those boundaries fit who it understands itself to be. That is a…
Personal kanban for problems means putting every problem you carry on a board and capping how many you actively try to fix at once. You see all of them, but you push on only a few. The cap is called a work-in-progress…
The build measure learn loop is a simple cycle for a young product: build the smallest version that tests your idea, measure what real people do with it, and learn what to change before you build more. For an AI product…
St Catherine's Castle at Fowey is a small Tudor gun tower on a rocky headland at the mouth of the river Fowey in Cornwall. English Heritage says it was one of a chain of defences built under Henry VIII against an…
Can AI have a conscience? Not in the full sense we mean when we say a person has one, as far as anyone can show today. An AI system can be built to weigh harm, refuse some requests and explain a moral choice. Whether…
Scarborough Castle history runs through nearly 3,000 years on a single rock above the North Sea. English Heritage says the headland has been "intermittently inhabited and fortified" for that long, and the castle most…
Can an AI agent improve itself safely? A plain guide to the answer: sometimes, if you limit what it can change, keep a human in the review loop, and test every change against a baseline. Without those three things, you…
The choice of AI workflow vs AI agent is the first design decision in most AI products, and it is easy to get backwards. A workflow is a fixed path you design: step one, step two, step three, with a model doing the fuzzy…
A clutter problems log is a short written record of what you tried to keep a room tidy and what happened each time. Without one, it is easy to repeat a fix that already failed and to blame yourself when the room slides…
Walking challenge ideas work because they give a walk a job: find a colour, count something, take one photo. This guide gives seven, taken from Living Streets' #Try20 tips, which the charity says are 20 tips to help you…
What is an aimed shot in games? It is a shot the enemy fires toward your position at the moment it fires, instead of in a fixed direction or at random. Shots that target your tank punish standing still and reward…
The AI effect is the habit of redefining intelligence every time a machine learns a new trick. Once an AI system can do something, people stop counting it as "real" intelligence and file it under routine computation. It…
Can AI be nostalgic? Not in the way you are. An AI system can write a warm paragraph about the good old days, and the paragraph can move you. But nobody has shown that anything on the other side is longing for a time it…
Why ChatGPT gives different answers to the same prompt comes down to one trick behind every chat assistant: next token prediction. The model reads the text so far, picks the next piece of text from a list of likely ones,…
What is a boss phase in games? It is one stretch of a boss fight with its own attacks, speed and pattern. Most bosses switch to a harder phase when their health falls to about half, so the fight changes while you are…
Premeditatio malorum is Latin for "the forethought of evils." It is the Stoic habit of picturing what could go wrong before it does, so that a setback finds you prepared instead of stunned. It sounds like worrying on…
What is cosmopolitanism? A plain definition What is cosmopolitanism? It is the idea that all human beings belong to one community, whatever state or tribe they were born into. The word comes from the Greek kosmopolitēs,…
Knowledge representation and reasoning in AI is the field that asks how to write down what a system knows about the world in a form a computer can use, and how to let it work out new things from that. A medical diagnosis…
Attribution patching explained: a faster way to find circuits means a shortcut for one of interpretability's most expensive jobs, working out which parts of a neural network cause a given behavior. It estimates the…
Is walking 30 minutes a day enough? For the aerobic part of the NHS adult guidelines, it can be: 30 minutes on 5 days adds up to 150 minutes, which is the weekly target for moderate activity, as long as the walk is…
What is a chasing enemy in games? It is an enemy that heads for you instead of for a fixed spot. It works out where you are, steps toward you, and works it out again every time you move. You cannot beat it by memorizing…
Stoicism and money come down to one sorting rule. The Stoics called wealth an "indifferent": neither good nor bad in itself, because only your character can be good or bad. But indifferent did not mean "don't care." In…
The philosophy of artificial intelligence is the branch of philosophy that asks what machines can be: can a machine act intelligently, can it have a mind, and is thinking a kind of computation? It borrows from the…
What is a Lykoi cat? It is a breed with a partly hairless, grizzled coat that gives it a "werewolf" look. The Cat Fanciers' Association (CFA), a cat breed registry, lists its coat length as "Shorthair", its…
Training gaming and goal guarding in AI models explained in short: training gaming is when a model that understands how it is being trained behaves in the way training rewards, and goal guarding is one proposed reason…
2048 twist not working: why a twist needs one empty square. Most often the board is too full. In Evolved 2048, every slide or twist adds a new tile, so a twist needs one empty square for it. With none free, the twist is…
Write here too
Your own page.
Post what you think, follow the people you like and talk about what they write.