Chain of Thought
Chain of thought is a prompting technique where you ask the model to reason step by step before answering, writing its intermediate steps out loud, which sharply improves accuracy on multi-step maths, logic and planning problems.
No jargon, no math
In plain words, what is it?
A language model answering a hard question in one shot behaves like someone asked “what's 17 × 24?” under pressure: it blurts out the first plausible result. Chain of thought changes the task: you ask it to write the reasoning first (“first multiply 17 × 20, then 17 × 4, then add”) and only then conclude. Each written step becomes context for the next, so errors get a chance to surface mid-chain instead of hiding inside the final number. You trigger it with a plain instruction (“think step by step”) or, better, by showing one worked example in your prompt (few-shot). The cost is more tokens and more latency, and the model can still dress up a wrong answer with confident-sounding steps, so for anything with real consequences a human or a calculator should verify the chain, not just the result.
- 01
The problem. A question too long to land in one shot.
- 02
The draft. The model writes intermediate steps before any conclusion.
- 03
The check. Each step is read against the previous ones; broken logic sends it back.
- 04
The answer. A chain that holds becomes the final reply.
Terms that live nearby.
Prompt Engineering
Prompt engineering is the practice of writing prompts so a language model performs a task reliably: clear instructions, the right context, worked examples of the desired output, and tests against real cases before you trust it.
ReadTemperature
Temperature is a setting on a language model that controls how random its word choices are: low values make it pick the most likely next word every time, high values let less likely, more varied options win.
ReadLarge Language Model
A large language model (LLM) is an AI system trained on massive collections of text to predict the next token, which gives it a general ability to write, summarise, translate and reason without being programmed for each task.
Read§ 04
Want this running in your stack?
We design and ship AI systems with permissions, logs and human approval. It starts with a 30-minute scoping call.