You won't believe what I discovered this week! 🤯 While diving into the latest AI developments, I stumbled upon something that literally made me jump out of my chair: a prompt that takes Gemini's reasoning abilities to a whole new level, making it think more like the sophisticated GPT-o1 model. And get this – it was created by a 17-year-old high school student! Meet Richards Tu, the brilliant mind behind what's now being called the "Thinking Gemini" prompt.
The Story Behind "Thinking Gemini"
This fascinating discovery has been making waves across the tech world. Several publications, including sina.com.hk, medium.com, and labs.botsnova.com, have been buzzing about this breakthrough. While the original prompt was designed for Claude, I've been experimenting with adapting these principles for Gemini, and the results are mind-blowing!
Want to try it yourself? The prompt is freely available here: Gist Link or scroll to the end of this article for the full prompt.
Breaking Down the Magic: How Does It Work?
Let me walk you through what makes this prompt so special. It's like giving Gemini a pair of glasses that helps it see problems in a whole new way:
-
Structured Thinking: Instead of just asking Gemini to spit out an answer, the prompt guides it through a thought process that feels incredibly human. It's like having a conversation with a brilliant friend who explains their reasoning step by step.
-
Step-by-Step Reasoning: Remember how your favorite math teacher would break down complex problems into bite-sized pieces? That's exactly what this prompt does for Gemini. No more mysterious leaps in logic!
-
Human-Like Thought Process: My favorite part? The prompt includes those little thinking moments we all have – the "Hmm..." and "Let me think about this..." that make the responses feel genuine and relatable.
-
Learning from Mistakes: Just like us humans, the prompt encourages Gemini to double-check its work and isn't afraid to say "Oops, let me correct that." This self-correction leads to much more reliable answers.
-
Balanced Thinking: It's like having both your analytical and creative brain working together – the prompt helps Gemini balance careful analysis with intuitive understanding.
Let's See It in Action!
I couldn't wait to put this to the test, so I tried it with the classic Wason Selection Task. The results honestly blew my mind! Here's what happened:
The Challenge: The Wason Selection Task
Let's put our enhanced Gemini to the test with a classic logic puzzle! Here's the scenario: Seven cards are placed on a table, each with a number on one side and a color on the other. The visible sides show: 50, 16, red, yellow, 23, green, and 30. The rule states: 'If a card shows a multiple of 4, then the opposite side is yellow.' Which cards must you turn over to test this rule?
Gemini's Response (Without Our Special Sauce):
Here's what regular Gemini suggested:
You need to turn over the following cards:
- 16: To check if the opposite side is yellow, as it's a multiple of 4.
- 50: To check that there is not a yellow side on this card.
- Red: To verify that there is not a multiple of 4 on the other side.
- Green: To verify that there is not a multiple of 4 on the other side.
- Yellow: To verify that there is not a multiple of 4 on the other side.
- 30: This card isn't needed because it doesn't fit in this logic.
The key is to check for both the cards that should be yellow according to the proposition and those cards that could violate it. 23 is also not needed.
Analysis: Interesting attempt, but there's some confusion here. Gemini is over-checking by including cards that can't actually violate the rule. It's like double-locking every door in your house when you only need to check the ones that might have been left open!
Now Watch What Happens With Our Enhanced Prompt:
Now this is where it gets exciting! With the prompt, Gemini walked through its thinking process like a master detective:
This is a classic logic puzzle, similar to the Wason Selection Task. The proposition is: "If a card is showing a multiple of 4, then the color on the opposite side is yellow." To test this, we need to consider both the cases where the condition (multiple of 4) is true and where it's false.
Let's break it down. The statement is a conditional ("If P, then Q"). To test a conditional statement, we need to check two things:
1. **Cases where P is true:** If a card shows a multiple of 4, then the opposite side *must* be yellow. So we need to check the cards showing multiples of 4.
2. **Cases where Q is false:** If a card shows a color that is *not* yellow, then the opposite side *cannot* be a multiple of 4. So we need to check the cards showing colors other than yellow.
Looking at the cards:
* **Multiples of 4:** 16 and 50 is not a multiple of 4. Therefore, only 16 needs to be checked.
* **Non-yellow colors:** Red, green. These cards need to be turned over to see if they have multiples of 4 on the other side.
Therefore, to test the proposition, we need to turn over the cards showing 16, red, and green. The card showing 50 is irrelevant because it doesn't show a multiple of 4. The yellow card is also irrelevant because it doesn't disprove the proposition, even if it had a multiple of 4 on the back. We are only interested in testing the cases that would falsify the statement.And boom! The perfect answer: turn over the cards showing 16, red, and green. No more, no less. Watching Gemini work through this was like seeing a master chess player explain their strategy!
Keeping It Real: What Are the Limitations?
Let's be honest – while this prompt is incredible, it's not a magic wand. Even with this clever prompt, Gemini sometimes stumbles on really complex problems or misses subtle nuances. It's like giving a calculator advanced features – it's still a calculator, not a mathematician. This reminds us that truly replicating human intelligence is still a fascinating challenge we're working to solve.
What's Next? The Future is Bright!
I'm genuinely excited about where this could lead! Here's what I think we should focus on:
- Testing the Limits: We need to really put this through its paces – try it with different types of problems and see where it shines (and where it doesn't).
- Making It Better: Could we tweak the prompt for specific types of questions? I bet we could!
- Spreading the Love: Could these same principles work with other AI models? I can't wait to find out!
This is just the beginning of a really exciting journey in making AI think more like us humans. While we're not quite at human-level reasoning yet, discoveries like this make me optimistic about the future.
I'd love to hear your thoughts and experiences if you try this out! Drop a comment below or reach out on social media. Let's explore this together! 🚀
This experiment was conducted using Gemini-1.5-flash