Google Gemini official thinking model tips and detailed case analysis

This article is translated from Googlet’s official account Google AI Developers, which shares Gemini thinking model tips and corresponding detailed cases:

1. Provide step-by-step instructions

Original Tip:

"Analyze the financial status of this company."

Improvement tips:

"Please analyze the financial situation of ABC Company and follow these steps:

• First, look at the revenue growth rate over the past 3 years and calculate the average
• Then, analyze profit margin trends, paying special attention to changes in gross and net profit margins
• Next, evaluate the company's debt-to-equity ratio and compare it to the industry average.
• Analyze cash flow, especially the ratio of operating cash flow to net profit
• Finally, based on the above analysis, an overall assessment of the company's financial health and a forecast for the next three years are given."

Why it’s more effective:

Clear steps guide the model through systematic thinking, ensuring that key financial indicators are not missed, and the analysis is carried out in a logical sequence to reach comprehensive conclusions.

2. Multiple sample prompts

Original Tip:

"Translate this sentence into French: 'I am glad to meet you'"

Improvement tips:

"Please translate the following sentence into French, following the style and formality of the examples I provided:

Example 1:
English: 'Hello, how are you today?'
French: 'Bonjour, comment allez-vous aujourd'hui?'
(used in formal business situations)

Example 2:
English: 'Hi there, what's up?'
French: 'Salut, ça va?'
(daily communication between friends)

I need the sentence translated: 'I am glad to meet you', please use formal French suitable for business situations.

Why it’s more effective:

By providing multiple translation examples and indicating the context, Gemini is able to understand the style and level of formality required, providing a more accurate and contextualized translation.

3. Define output and behavior

Original Tip:

"Write me a resume"

Improvement tips:

"System Instructions: You are a professional resume consultant who specializes in creating attractive resumes for people in the technology industry.

I need you to create a one-page resume for an engineer with 5 years of software development experience, focusing on front-end development. Please follow the following behavior and output requirements:

• Divide your resume into 5 sections: Personal Profile, Skills Summary, Work Experience, Education, and Personal Projects
• Each work experience should include 3-4 descriptions of achievements, using quantitative data
• Start each description with an action-oriented verb
• The final output should be well formatted, easy to read and no longer than 500 words
• Before responding, please verify that your resume contains all required sections and that the description is specific enough.

Personal information: Zhang Ming, Bachelor of Computer Science from Fudan University, worked as a front-end developer at Alibaba and Xiaomi, and is proficient in React, Angular, and Vue.js frameworks.

Why it’s more effective:

It clearly defines the role the model should play, the format and content requirements of the output, and requires the model to verify itself to ensure that the output meets the required quality standards.

4. Debugging and Booting

Original Tip:

"Explaining Quantum Entanglement"

Question reply:

[Assuming Gemini provides an explanation, but confuses the concepts of quantum entanglement and quantum superposition]

Debug and Boot:

"I noticed that you mixed the concepts of quantum entanglement and quantum superposition in your explanation. Let's adjust them:

• First, please clearly distinguish between quantum entanglement and quantum superposition, two different quantum phenomena.
• Then, we focus on explaining quantum entanglement, specifically what Einstein called 'spooky action at a distance'
• Please explain in detail how when two particles are entangled, measuring one particle will immediately affect the state of the other particle, no matter how far apart they are.
• Finally, let's take a simple experimental example to illustrate quantum entanglement, such as the EPR paradox or Bell's inequality experiment.

Make sure your explanations are physically accurate, but try to use language that the average person can understand."

Why it’s more effective:

By pointing out specific conceptual confusions and providing clear directions for correction, the model is guided to reorganize its thinking, ensuring that the next answer is more accurate and organized.

5. Make full use of long output and long thinking

Original Tip:

"Write me an article about climate change"

Improvement tips:

Please write an in-depth analysis of the impact of climate change on global food security. Please write an article of 3,000-4,000 words.

Before you start writing, please:

• List 5 major climate change factors you will discuss (e.g. increase in extreme weather events, changes in precipitation patterns, etc.)
• Determine how these factors specifically affect agricultural production in different regions
• Consider potential adaptation and mitigation strategies

In the final article I need to:

• Introduction: Outline the problem and structure of the article
• Detailed analysis of each factor, including scientific data support
• Comparison of differential impacts in different regions (Africa, Asia, North America, etc.)
• Prediction of food security trends in the next 30 years
• Science-based recommendations for mitigation and adaptation strategies
• Conclusion: Synthesis and Call to Action

Please use data from the latest climate science and IPCC reports in your article. Make sure your analysis has both breadth and depth, and avoid oversimplifying complex issues. "

Why it’s more effective:

By specifying in detail the content structure, required depth of analysis, data sources, and specific length requirements, this prompt guides the model to think deeply and systematically and produce a comprehensive, insightful long-form article rather than a brief overview that talks about generalities.

6. Additional example: Structured prompt format (from user @GoKenja)

Example of structured prompt:

Objective: Create a sci-fi story outline, set in a post-climate disaster world, where the protagonist discovers a technology that can restore the ecosystem Instructions: - Create a story outline with a three-act structure - Include the protagonist, antagonist, and 3-4 supporting characters - Design a core moral dilemma - Make sure the scientific elements are sufficiently believable - The outline length should be between 800-1000 words Background: The story is set in 2150, the global temperature has risen by 4 degrees, most coastal cities have been submerged, and humans live in cities or inland self-sufficient communities within high-tech protective shields. The protagonist is a biotechnology researcher. Sample Output: # [Title of the Story] ## Act I: Setup & Challenge [Briefly summarize the main events of Act I...] ## Act II: Conflict & Exploration [Briefly summarize the main events of Act II...] ## Act III: Resolution & Transformation [Briefly summarize the main events of Act III...] ## Characters: - Protagonist: [name], [brief description] - Antagonist: [name], [brief description] - Supporting Characters: [List 3-4 characters and their brief descriptions] ## Core Science Concepts: [Explain the core scientific elements of the story and their plausible basis] ## Moral Dilemma: [Describe the core moral issue in the story]

Why it’s more effective:

This structured prompt clarifies the creation goal, specific instructions, necessary context, and expected output format, providing the model with a clear thinking framework to ensure that the output is both creative and meets specific needs. With the complete structure of goals, instructions, context, and output examples, the model can more fully understand the task requirements and provide high-quality responses.