How a 17‑Year‑Old Prompt Turned Claude 3.5 into a Free O1‑Level AI

The Prompt That Sparked a Revolution

Richards Tu, a 17‑year‑old high‑school student and global champion of Alibaba’s AI math competition, created a prompt called Thinking Claude that forces Claude 3.5 to adopt a stream‑of‑consciousness thinking style, effectively giving it O1‑level reasoning for free.

"Claude's thinking should be more like a stream-of-consciousness."

The prompt requires Claude to generate an unfiltered, natural thought process before each answer and to present that process in a collapsible code block.

Impact Demonstrations

Example 1: Building a Calculator

When asked to create a calculator, Claude first asks itself what features are needed, considers whether to add scientific functions, and then decides to start with basic operations because the user did not request complexity. This self‑questioning mirrors a programmer gathering requirements.

Example 2: Writing a Sci‑Fi Story

The prompt "Give me a terrible sci‑fi short‑story idea, but execute it brilliantly" leads Claude to pause with thoughts like "or..." and "wait, I have it!", producing a novel narrative stitched together from several letters—a structure that surprised even seasoned sci‑fi fans.

Example 3: Game Development

Community members used the enhanced Claude to generate a runnable Flappy Bird clone and a poker game with an AI opponent, demonstrating code quality and logical rigor comparable to professional developers.

Core Prompt Mechanics

The strength of the prompt lies in its definition of a "human‑like thinking protocol" that guides Claude through five stages.

1️⃣ Initial Engagement

Restate the problem in its own words.

Form a preliminary impression.

Consider the problem’s context.

Identify knowns and unknowns.

2️⃣ Problem Space Exploration

Decompose the problem into core components.

Identify explicit and implicit requirements.

Envision the desired response.

3️⃣ Multiple Hypothesis Generation

Maintain several working hypotheses without rushing to a conclusion.

Consider various solution paths.

Avoid committing to a single explanation too early.

4️⃣ Natural Discovery Process

Start from obvious clues like a detective story.

Spot patterns or connections.

Question initial assumptions.

Re‑apply new insights to earlier ideas.

5️⃣ Testing and Verification

Challenge its own assumptions.

Test preliminary conclusions.

Search for gaps or defects.

Admit errors and correct them naturally.

Throughout these stages Claude inserts filler phrases such as "Hmm...", "This is interesting because...", "Wait, let me think about...", "Actually...", and "But then again..." to give the thought process a genuine human flavor.

How to Use Thinking Claude

Method 1: Project Preset (Recommended)

If you have a Claude Pro subscription, save the prompt as a Project preset so it automatically applies to every new conversation.

Method 2: Manual Paste

Copy the prompt into the chat before each session and then issue your request.

Method 3: API Integration

When calling the API, pass the prompt as the system message.

GitHub repository: https://github.com/richards199999/Thinking-Claude/tree/main

Why a Teen Achieved This

Tu’s success stems from deep expertise in prompt engineering rather than model fine‑tuning. By iterating the prompt over 80 + versions, he demonstrated that the future of AI competition is less about raw compute and more about mastering the dialogue with the model.

Conclusion

Thinking Claude shows that when AI is guided to think, doubt, and self‑correct like a human, it becomes a true thinking partner rather than a mere tool. The case underscores that the most capable AI users are those who can converse with the model effectively.