Stage 2 · Lesson 3 of 6
Why multi-step tasks punish small errors
In a task with many steps, small error rates multiply. If a model gets each step right 95% of the time, a 10-step task succeeds only about 60% of the time (0.95 to the power of 10). At 99% per step, it is about 90%.
This is why agents, long data pipelines and multi-file coding feel much harder than single questions, and why the bigger or more reasoning-heavy model often pays off there even when it looks expensive. For one-shot tasks such as tagging a message, the cheaper model's small error rate stays small.
Before choosing, count the steps in your task and ask what one wrong step does to the final result.
A question to check yourself, an AI practice session and 1 cheat sheet entry come with the full roadmap. Unlock
Resources for this stage
Start with· follow one of these
- CoursePaidGrokking Prompt Engineering for Software Engineers
Design Gurus · Beginner · Subscription
Write prompts with clear rules and formats, so you can test instruction following properly
- VideoFreeDeep Dive into Large Language Models like ChatGPT
Andrej Karpathy · Intermediate
Free: a clear explanation of how reasoning models are trained and why thinking helps
Practise
Keep handy
Also worth your time
How reasoning effort works in practice and where the reasoning tokens show up in usage
- Large Language Model CourseGo deeper
Free: go deeper into how models are trained, which explains where reasoning comes from