Guided prompt engineering
The Latent Space Explorer
A visual metaphor for how better prompts narrow ambiguity. Step through task context, constraints, examples, failure policy, and tradeoffs to see why each addition changes the likely answer.
Guided Levels
Concept Map
Embedding Vector
Picture 512 dimensions as coordinates
A model does not place text in only x, y, and z. It represents meaning as hundreds or thousands of numeric coordinates. This simplified 512 dimensional vector shows many tiny signals that are projected down for the 3D scene above.
Current Lesson
Level 1: Vague Request
Prompt
> Write a program for a website.
Nothing is pinned down yet — every detail is left for the model to guess.
Why it improves
This view uses latent space as a metaphor: the request points at many possible patterns, so the response drifts toward generic web examples.
Still missing
The model has no product goal, framework, user, constraints, quality bar, or definition of done.
Next prompt move
Name the task, the target user, and the implementation context before asking for code.
Likely output shape
<html>
<body>
<h1>Hello World</h1>
<script>alert('Welcome');</script>
</body>
</html>