16 posts found
What a large language model actually predicts
A model does not look anything up and does not decide what is true. It estimates which token comes next. Almost everythi...
Tokens, not words: how a model reads your text
Models do not see characters or words. They see tokens — and once you know how text becomes tokens, several odd behaviou...
Why temperature changes the answer, not the knowledge
Turning temperature down does not make a model more accurate. It makes it more repeatable — and confusing the two is how...
The context window is a budget, not a memory
Bigger context windows did not give models memory. They gave you a larger envelope to fill on every single request — and...
Why models hallucinate, and what actually reduces it
Hallucination is not a glitch that a better model will one day remove. It is what generation does when it has nothing to...
Choosing between a large model and a small one
Most production traffic does not need the largest model available. Routing by task rather than defaulting to the top of...
Why agents fail on long tasks
An agent that handles five steps beautifully can fall apart at twenty. The reason is rarely reasoning — it is that the c...
When a workflow beats an agent
If you already know the steps, do not ask a model to rediscover them on every request. A fixed pipeline with model calls...
Reviewing AI-generated UI: a checklist
Generated interfaces fail in a predictable set of places. Knowing which ones turns review from a vague unease into a lis...
Prompt caching: the cheapest speedup you are not using
If every request begins with the same two thousand tokens of instructions, you are paying full price to resend them. Ord...
Streaming responses without breaking your UI
Streaming makes a slow response feel fast. It also means rendering text that is syntactically incomplete, and handling a...
How to evaluate an AI feature before you ship it
You cannot unit test "is this a good answer", but you can build a set of real cases with known-good outputs. An afternoo...