SecondBrain
Ask the Brain
Index/Conceptupdated Sat May 30 2026 08:00:00 GMT+0800 (Philippine Standard Time)

Intelligent Hill (Prompting Camps)

promptingai-literacythemitmonk-frameworkprompt-engineering

Intelligent Hill (Prompting Camps)

Sandeep Swadia's metaphor for the prompting hierarchy, from Dangerously Smart with AI (theMITmonk). The "hill" is the climb from naive prompting to agent-level work; each "camp" is a more sophisticated technique.

Reframes the LLM from calculator to probability engine: "If you ask the same question to AI again, it'll give you a completely different answer. It'll happily make things up for you unless you ask it to verify." So architecting the prompt matters.

The five camps

Camp 0 — Zero-shot (what most people do)

"Give me the best new business idea" — no examples, no context. "You're literally rolling the dice and looking to win."

Camp 1 — One-shot prompting

One clear example as a style anchor:

"Write a LinkedIn post about remote work. Use this specific post as a style guide" + pasted reference post.

"That simple act is already an upgrade than rolling the dice blindly."

Camp 2 — Few-shot prompting

3+ examples so the model finds patterns of style, substance, tone. Sandeep calls this "grounding the model."

"Here are five of my previous presentations and now write a new presentation based on my tone of voice on topic XYZ."

Pro tip from the video: "Ask the AI to explain the pattern back to you first. That way AI is forced to articulate what it's doing and more importantly, you're forced to learn how your brain works."

Camp 3 — Chain-of-thought reasoning

Make the model show its work. "Your job is to slow AI down. Enforce explicit clarity by asking it to show its work." Also reduces hallucinations.

Worked example:

"Do not refine my research report yet. List the top three most impactful areas of improvement after we analyze it. Tell me why you think so and suggest how we address each. Think step-by-step. Show me your thinking for each step."

Sandeep: "That last line is the most important one."

Camp 4 — Agents

"The best way to think about agents is to think about who you would hire for a task."

Single-prompt multi-role task assignment:

"Do deep research on trends on topic XYZ, analyze and cross-reference all the trends to find the three most important ones, and draft a one-page memo summarizing the findings."

(Researcher + analyst + copywriter packed into one prompt — agent decides how to allocate.)

The actionable challenge

"Open your favorite AI app and take any prompt that you were about to use. Just try to get to the next camp."

Pithy frame: "When you are dealing with a drunk genius, make sure you're the one driving the car."

How this fits into the vault

  • First explicit prompting taxonomy in this vault. The wiki has PRIME Framework (forthcoming, structural) and AIM Protocol (lightweight, drafting-specific), but the canonical zero/one/few-shot + CoT + agents hierarchy hadn't been named.
  • Connects to You're Not Behind (Yet) Learn AI Agents (theMITmonk) — Camp 4 (Agents) is where Sandeep picks up the thread in his later agents-focused video, with ARR / OODA / GPS.
  • Connects to LLM as Judge — Camp 3's "show your work" is one of the standard hooks for LLM-as-judge evaluation pipelines.
  • Note on naming: "zero-shot / one-shot / few-shot" are standard prompt-engineering terms (not Sandeep's coinage); the "camp" metaphor is his. Page name uses the metaphor because it's the unique framing this video contributes.

Sources