Acadbyte
Browse the library

Prompt Engineering

The Killer 15, ranked simple to advanced

Fifteen techniques that survive contact with real work, each one built the same way: why it exists, the template you copy, a worked example from tech, a worked example from ordinary life, and a board you play until the idea sticks.

Beginner–Advanced180 min15 topics

One topic plays without an account — the first one you start, anywhere in the library. A free account opens the rest and keeps everything you have already done.

0 of 15 topics

Beginner · The Everyday Five

The five you will reach for most often. Least effort, and the most rewriting saved.

  1. Contextual Zero-Shot PromptingBeginnerThirty extra seconds of typing changes the answer more than any model upgrade will. The strange part is which thirty seconds.12 min
  2. RTF Framework (Role, Task, Format)BeginnerThree slots, and almost everybody leaves the same one empty. It is also the one that costs them the most rewriting.12 minComes after Contextual Zero-Shot Prompting
  3. Constraint PromptingBeginnerHalf of what comes back is padding nobody asked for. Adding more instructions makes it worse, and the fix runs in the opposite direction.12 minComes after Contextual Zero-Shot Prompting
  4. Persona AssignmentBeginnerOne reasoning task climbed from 23.8% to 84.2% accuracy. The model did not change, and neither did the question.12 minComes after Contextual Zero-Shot Prompting
  5. Few-Shot PromptingBeginnerA paragraph describing the style you want is the lossy version of the request. The precise version contains no adjectives at all.12 minComes after Contextual Zero-Shot Prompting

Intermediate · Professional Grade

Techniques that earn their place once the task has real structure and the first draft has to hold up.

  1. COSTAR FrameworkIntermediateSix letters, and one of them is the reason your last proposal came back reading like a brochure. Most people fill that line in with the wrong word.12 minComes after Contextual Zero-Shot Prompting
  2. Chain-of-Thought PromptingIntermediateAsk for the answer first and the explanation second, and you get something that reads more convincing than the correct version. That is the trap.12 minComes after Contextual Zero-Shot Prompting
  3. Flipped PromptingIntermediateEvery technique so far assumes you know what the model needs to be told. This one is for the much more common case where you do not.12 minComes after Contextual Zero-Shot Prompting
  4. Iterative Prompt ChainingIntermediateRead the output of a very large request from the bottom up. Section six is always thinner than section one, and the reason has nothing to do with how good your prompt was.12 minComes after Contextual Zero-Shot Prompting
  5. Self-Critique and RefinementIntermediateAsk a model to make its own draft better and it hands you a different average draft. Ask it one specific question about that draft and it finds the hole you missed.12 minComes after Contextual Zero-Shot Prompting
  6. Negative Space PromptingIntermediateAsk a model whether your plan is good and it will tell you your plan is good. The uncomfortable part is that you already knew that when you asked.12 minComes after Contextual Zero-Shot Prompting

Advanced · High Stakes

For work where being confidently wrong is expensive: verification, branching, and prompts that write prompts.

  1. Tree of Thought PromptingAdvancedThree answers instead of one sounds like three times the work for three times the value. It is neither, and the difference is one line most people leave out.12 minComes after Contextual Zero-Shot Prompting
  2. Chain of Verification (CoVe)AdvancedA model can write a false sentence and also tell you it is false. It just cannot do both in the same breath, which turns out to be the entire opportunity.12 minComes after Contextual Zero-Shot Prompting
  3. Meta-PromptingAdvancedThe best prompt for a task you have never done before is one you are not qualified to write. That turns out not to be a problem.12 minComes after Contextual Zero-Shot Prompting
  4. Agentic Prompt OrchestrationAdvancedFour prompts in a row is not a pipeline. One line separates the two, and its absence does not show up until the last step, if it shows up at all.12 minComes after Contextual Zero-Shot Prompting
All tracks