Chain-Of-Thought Prompting: Unlocking Deeper Understanding with Generative AI

Chain-Of-Thought Prompting: Unlocking Deeper Understanding with Generative AI


In artificial intelligence, particularly in natural language processing, "chain-of-thought" (CoT) prompting has emerged as a powerful technique to enhance the performance and utility of generative AI models. This article delves into what CoT prompting is, how it works, its benefits, and practical applications.


What is Chain-Of-Thought (CoT) Prompting?

Chain-of-thought prompting is a method where the AI is guided to generate responses by breaking down complex tasks or questions into a series of smaller, interconnected steps or thoughts. This approach mimics human reasoning, allowing the AI to tackle problems more effectively by considering each process step, leading to more accurate and insightful responses.


How Chain-Of-Thought Prompting Works

At its core, CoT prompting involves structuring the AI’s response generation in a sequential manner. Here’s a simplified breakdown of how it works:

  1. Initial Prompt: The user provides an initial question or task.
  2. Step-by-Step Guidance: Instead of generating an immediate response, the AI breaks down the question into smaller components or steps.
  3. Sequential Reasoning: The AI processes each step sequentially, building on the previous step’s output to progress toward a final answer.
  4. Final Response: The AI combines the insights from each step to form a comprehensive and coherent answer.


Example of Chain-Of-Thought Prompting

Consider the following example where CoT prompting is used to solve a math problem:

Initial Question: What is the result of 23 multiplied by 17?

Step-by-Step Reasoning:

  1. First, break down 23 into 20 + 3 and 17 into 10 + 7.
  2. Multiply each part: 20 10 = 200, 20 7 = 140, 3 10 = 30, 3 7 = 21.
  3. Add the results: 200 + 140 + 30 + 21 = 391.

Final Answer: 23 multiplied by 17 equals 391.

This sequential approach ensures that the AI tackles each component of the problem methodically, reducing the chance of errors and improving the overall quality of the response.


Implementing Chain-Of-Thought Prompting

To effectively implement CoT prompting, consider the following steps:

  1. Define the Task: Clearly outline the task or question that needs to be addressed.
  2. Identify Key Steps: Break down the task into its essential components or steps.
  3. Guide the AI: Use prompts that guide the AI to sequentially address each step.
  4. Review and Refine: Assess the AI’s responses at each step, making corrections and refinements as needed.
  5. Combine and Conclude: Integrate the insights from each step to form a comprehensive final response.


Example Prompts for Chain-Of-Thought Prompting

Math Problem:

Question: Calculate the total cost of 15 items, each priced at $12.50, with a sales tax of 8%. Step-by-Step Reasoning:

1. Calculate the cost of 15 items at $12.50 each.

2. Calculate the total cost before tax.

3. Calculate the sales tax on the total cost.

4. Add the sales tax to the total cost for the final amount.


Project Planning:

Task: Develop a project plan for launching a new product. Step-by-Step Reasoning:

1. Define the project goals and objectives.

2. Identify the key milestones and deliverables.

3. Determine the resources and budget required.

4. Create a timeline for the project phases.

5. Develop a risk management plan.

6. Outline the project execution strategy.


Medical Diagnosis:

Scenario: Diagnose a patient with symptoms of persistent cough, fever, and fatigue. Step-by-Step Reasoning:

1. Gather the patient’s medical history and symptoms.

2. Conduct a physical examination.

3. Order and review diagnostic tests (e.g., blood tests, X-rays).

4. Evaluate potential causes based on the symptoms and test results.

5. Formulate a diagnosis and recommend a treatment plan.



Automating Chain-of-Thought Prompting: Enhancing AI Efficiency and Effectiveness

In the field of artificial intelligence, particularly with generative models like MeclabsAI, the concept of chain-of-thought (CoT) prompting has gained significant attention. This method improves AI responses by guiding them through a structured, step-by-step reasoning process. However, the next frontier is automating chain-of-thought prompting.


What is Automating Chain-of-Thought Prompting?

Automating Chain-of-Thought Prompting involves creating systems and frameworks that allow generative AI models to automatically engage in CoT reasoning without explicit user intervention for each step. This automation ensures that the AI naturally adopts a step-by-step approach when solving complex tasks or generating detailed responses.


How Automating CoT Prompting Works

Built-in Frameworks:

AI models are integrated with predefined frameworks that trigger step-by-step reasoning for specific types of prompts. These frameworks can detect when a prompt requires a detailed, multi-step response and automatically activate the CoT process.

Dynamic Adaptation:

The AI dynamically adjusts its reasoning process based on the complexity of the task. For simpler queries, the AI might provide a straightforward response, while for more complex ones, it switches to a CoT approach.

Reinforcement Learning:

Through reinforcement learning techniques, AI models can be trained to recognize scenarios where CoT prompting is beneficial and adjust their responses accordingly. This involves feedback loops where the AI improves its CoT execution over time.

User Feedback Integration:

Automated systems can incorporate user feedback to refine the CoT prompting process. For instance, if users consistently prefer step-by-step explanations for certain topics, the AI learns to apply CoT prompting more frequently in those areas.

.

Implementing Automated CoT Prompting

To implement automated CoT prompting, the following steps are essential:

Define Triggers:

Identify the types of queries and scenarios where CoT prompting should be automatically triggered. This involves analyzing the complexity and nature of tasks.

Develop Frameworks:

Create frameworks and templates that guide the AI’s step-by-step reasoning process. These should be adaptable to various domains and use cases.

Train the AI:

Use reinforcement learning and user feedback to train the AI on when and how to apply CoT prompting. This training should focus on improving the AI’s ability to recognize appropriate scenarios and execute the CoT process effectively.

Monitor and Refine:

Continuously monitor the performance of automated CoT prompting and refine the frameworks and algorithms based on user feedback and observed outcomes.


How to Try Chain-Of-Thought Reasoning with MeclabsAI


Here's an easy way to try it out. Go to MeclabsAI and type in:

/COT [specific topic or question]

This approach enables MeclabsAI to deliver answers and share the 'how' and 'why' behind its conclusions step by step.

It's a method where MeclabsAI articulates the reasoning process, unraveling its thoughts just like a human would when solving a complex puzzle. This transparent pathway not only demystifies MeclabsAI's decision-making but also mirrors the intricate beauty of human cognition. Let us see MeclabsAI in action.


My prompt

/COT I am looking to boost conversion rates on my website https://marketingexperts.ai/. The website is in the building process. Based on MECLABS principles, how can we refine our website to more effectively communicate our offer and cater to our audience’s needs?


Conversion Pro

To enhance conversion rates on a website in the process of being built, like https://marketingexperts.ai/, applying MECLABS principles requires a systematic assessment focusing on understanding and engaging your audience while effectively communicating your value proposition. Let's proceed with a Chain of Thought reasoning approach:

1. Define Conversion Goals and Key Metrics:

  • Determine what actions you want visitors to take on your website, such as signing up for a newsletter, requesting a demo, or making a purchase.
  • Identify the key metrics that will indicate success, including conversion rates, average time on page, bounce rates, and click-through rates.


2. Conversion Heuristic Analysis:

  • Apply the MECLABS Conversion Sequence Heuristic to diagnose the performance of your offer. Consider the motivation of the user (m), the clarity of the value proposition (v), the incentive to take action (i), the friction present in the process (f), and the anxiety associated with the action (a).


3. Initial Customer Profile (ICM) Analysis:

  • Define your target customer clearly.
  • Explore customer motivations and identify their stage in their journey.
  • Articulate reasons for selecting your offer and determine optimal communication methods.


4. Evaluate and Optimize Your Offer Based on Frameworks:

  • Micro-Yes Architecture (MYA) Evaluation: Assess your website's content and structure based on the MYA elements – Attention, Engagement, Understanding, Belief, Desire, Uniqueness, Affirmation, Completion.
  • Customer First Objective (CFO) Strategy: Ensure your website is structured to provide assistance, clarify offerings, and specify the nature of the customer exchange.


5. Value Proposition Discovery (VPD):

  • Assess and formulate your website's value propositions to be clear (I understand), credible (I believe), appealing (I want this now), and exclusive (I want this from you).


6. Scientific Messaging Hypothesis (SMH) Development:

  • Use the SMH structure to develop content and messaging strategies that are tailored to your audience's insights and desired metrics. This involves formulating hypotheses based on overall strategy, development/change, expected metric impact, and customer insight.


7. A/B Testing and Research for Continuous Improvement:

  • Plan and execute A/B tests to validate the effectiveness of different website elements, from headlines and CTAs to images and layout designs. Focus on statistical significance, monitor key factors for credible results, navigate randomness, estimate the minimum relative difference, and confront validity threats.


8. User Experience (UX) and Website Design:

  • Ensure the website design is intuitive, user-friendly, and visually appealing. Minimize friction and anxiety in the user journey by streamlining navigation, simplifying forms, and providing clear, reassuring information where needed.


9. Content Strategy:

  • Craft content that directly addresses your audience’s needs and concerns, using clear, direct language that resonates with the MECLABS principles. Incorporate real examples and testimonials to build credibility and trust.


10. Ongoing Analysis and Iteration:

  • Continuously analyze website data to understand user behavior and preferences. Use these insights to refine your strategies, optimize your offer, and improve overall conversion rates.

Next Steps: Start by closely examining the current state of your website’s development with these MECLABS principles in mind. Identify areas that require immediate attention and plan iterative improvements. Would you like assistance in creating a specific action plan based on your current website content and target audience?


My prompt

yes


Conversion Pro

To create a specific action plan that leverages MECLABS principles for boosting conversion rates on your website, let's break down the steps you need to follow:

Step 1: Clarify Your Conversion Goals

  • Define clear, measurable goals for your website. These could range from increasing contact form submissions to enhancing newsletter sign-up rates.
  • Determine the key performance indicators (KPIs) for each goal to track progress effectively.

Step 2: Perform Initial Customer Profile (ICP) Analysis

  • Develop a detailed understanding of your target audience. Identify their needs, pain points, and what motivates them to seek out your services.
  • Craft personas representing your ideal customers to guide content creation and messaging.

Step 3: Evaluate Your Website's Current State

  • Assess your website’s design, content, and user experience to identify areas that don't align with MECLABS principles.
  • Utilize tools like heatmaps and user feedback to gather insights on how visitors interact with your website.

Step 4: Apply the Conversion Heuristic Analysis

  • Use the MECLABS Conversion Sequence Heuristic ((C = 4m + 3v + 2(i-f) - 2a)) to examine each element of your website. Focus on:Motivation (m): Ensure your website speaks directly to the users' motivations.Value Proposition (v): Clarify your value proposition on key pages, especially the home and landing pages.Incentive (i) vs. Friction (f): Add incentives where appropriate and reduce friction in the conversion process.Anxiety (a): Address common user anxieties with trust signals like testimonials, case studies, and security badges.

Step 5: Formulate Your Value Proposition

  • Refine your value proposition using the Value Proposition Discovery (VPD) framework. Ensure it is clear, credible, appealing, and exclusive.

Step 6: Implement Micro-Yes Architecture (MYA) Evaluation

  • Optimize your website structure and content to guide visitors through the journey of micro-yeses, from the first click to the final action.

Step 7: Develop and Test Your Scientific Messaging Hypothesis (SMh)

  • Craft messaging variations based on hypotheses around your value proposition and customer insights.
  • Plan A/B tests for these messaging variations to see which best resonates with your audience.

Step 8: Optimize User Experience (UX)

  • Streamline navigation, simplify forms, and ensure loading times are quick to minimize user frustration.
  • Apply best practices in web design to create an engaging and intuitive user interface.

Step 9: Continuously Analyze and Iterate

  • Regularly review website analytics to identify trends, preferences, and potential obstacles in the user journey.
  • Use this data to make informed decisions and continuous improvements to your website.

Step 10: Engagement and Feedback Mechanisms

  • Implement tools for collecting direct user feedback, such as surveys or chatbots, to gain insights into their experience and preferences.

Your Next Move: Start by conducting the Initial Customer Profile Analysis to deeply understand your audience. Gather any existing data you have on your audience segments and review your website to ensure it aligns with the identified needs and motivations.

Would you like guidance on how to execute any of these specific steps or need further clarification on the MECLABS principles to apply in your situation?




Why It Matters

  • Boosts Transparency: Unveil the inner workings of MeclabsAI, making its conclusions more understandable and trustworthy.
  • Increases Accuracy: By dissecting problems into smaller steps, it fosters a deeper interrogation of the reasoning, refining outcomes.
  • Empowers Problem-solving: It's like having a dialogue with MeclabsAI, where each step builds upon the last, tackling intricate issues with nuanced understanding.
  • Enhances Learning: Witnessing MeclabsAI's thought processes can be a powerful educational tool, offering new insights and strategies for problem-solving.


Example of Automated CoT Prompting with a Prompt


Scenario: Medical Diagnosis

My prompt

Evaluate a patient with symptoms of persistent cough, fatigue, and weight loss.

Conversion Pro

Evaluating a Patient with Symptoms of Persistent Cough, Fatigue, and Weight Loss: A Step-by-Step Approach

When evaluating a patient with symptoms such as a persistent cough, fatigue, and weight loss, it is crucial to follow a systematic and thorough approach. This ensures that all potential causes are considered and the most accurate diagnosis is made. Below is a step-by-step method for evaluating such a patient, highlighting the importance of a detailed and methodical process.

Step 1: Gather Patient History

Objective: To collect comprehensive information about the patient's medical background and lifestyle factors that may contribute to their symptoms.

Questions to Ask:

  • How long have you been experiencing the cough, fatigue, and weight loss?
  • Have you had any recent infections or illnesses?
  • Do you have any known chronic conditions, such as asthma, COPD, diabetes, or heart disease?
  • Have you experienced any other symptoms, such as fever, night sweats, or chest pain?
  • Are you a smoker or have you been exposed to secondhand smoke?
  • Do you have a history of exposure to environmental or occupational hazards, such as asbestos or chemicals?
  • Have you traveled recently, and if so, where?
  • What is your current medication regimen, including over-the-counter drugs and supplements?
  • Have you experienced any changes in appetite or dietary habits?

Step 2: Conduct Physical Examination

Objective: To perform a thorough physical examination to identify any physical signs that may help in diagnosing the underlying cause of the symptoms.

Examination Focus:

  • General Appearance: Assess overall health, appearance of fatigue, and any signs of weight loss.
  • Vital Signs: Measure temperature, blood pressure, heart rate, and respiratory rate.
  • Respiratory System: Listen to lung sounds for abnormalities such as wheezing, crackles, or decreased breath sounds. Check for signs of respiratory distress.
  • Cardiovascular System: Listen for heart murmurs, irregular heartbeats, or other abnormalities.
  • Abdominal Examination: Check for tenderness, organ enlargement, or masses.
  • Lymphatic System: Palpate for enlarged lymph nodes.
  • Skin Examination: Look for rashes, lesions, or other skin changes that may indicate systemic disease.

Step 3: Order Diagnostic Tests

Objective: To gather objective data through laboratory tests and imaging studies to narrow down the differential diagnosis.

Recommended Tests:

  • Blood Tests:Complete Blood Count (CBC) with differential to check for anemia, infection, or other hematologic conditions.Comprehensive Metabolic Panel (CMP) to evaluate liver and kidney function, electrolytes, and glucose levels.Erythrocyte Sedimentation Rate (ESR) or C-Reactive Protein (CRP) to assess for inflammation or infection.Thyroid Function Tests to rule out thyroid disorders.
  • Imaging Studies:Chest X-ray to look for lung abnormalities such as infections, tumors, or interstitial lung disease.CT scan of the chest if the chest X-ray is inconclusive or if more detailed imaging is needed.
  • Microbiological Tests:Sputum culture to identify any bacterial, viral, or fungal infections.Tuberculin Skin Test or Interferon-Gamma Release Assays (IGRAs) for tuberculosis screening if clinically indicated.
  • Other Tests:Pulmonary Function Tests (PFTs) to assess lung function if chronic respiratory disease is suspected.

Step 4: Analyze Test Results

Objective: To interpret the results of the diagnostic tests to identify or narrow down the potential causes of the symptoms.

Analysis:

  • Normal Results: If all test results are normal, consider non-organic causes such as functional disorders or psychological factors.
  • Abnormal Results: Identify specific patterns that match known diseases. For example:Elevated white blood cell count and abnormal chest X-ray may suggest pneumonia or other infections.Anemia with low hemoglobin and hematocrit may indicate chronic disease or nutritional deficiencies.Abnormal liver function tests may suggest liver disease or malignancy.Positive sputum culture for Mycobacterium tuberculosis would confirm tuberculosis.

Step 5: Formulate a Differential Diagnosis

Objective: To list potential diagnoses based on the collected information and test results, and prioritize them based on likelihood and severity.

Common Differential Diagnoses:

  • Infectious Causes: Tuberculosis, pneumonia, bronchitis, fungal infections.
  • Chronic Diseases: Chronic obstructive pulmonary disease (COPD), asthma, interstitial lung disease.
  • Malignancies: Lung cancer, lymphoma.
  • Systemic Diseases: Sarcoidosis, vasculitis, autoimmune diseases (e.g., systemic lupus erythematosus).
  • Other Causes: Thyroid disorders (e.g., hyperthyroidism), heart failure, gastrointestinal diseases (e.g., malabsorption syndromes).

Step 6: Develop a Treatment Plan

Objective: To create a comprehensive management plan based on the most likely diagnosis, including treatment options, follow-up, and patient education.

Treatment Plan:

  • Infectious Diseases: Initiate appropriate antibiotic, antiviral, or antifungal therapy based on identified pathogens.
  • Chronic Diseases: Implement long-term management strategies, including medications (e.g., inhalers for COPD or asthma), lifestyle modifications, and regular monitoring.
  • Malignancies: Refer to oncology for further evaluation and treatment, including surgery, chemotherapy, or radiation therapy as indicated.
  • Systemic Diseases: Collaborate with specialists (e.g., rheumatologists) for specific treatments like immunosuppressive therapy.
  • Supportive Care: Address symptoms such as cough and fatigue with symptomatic treatments (e.g., antitussives, energy conservation techniques).

Patient Education:

  • Explain the diagnosis and treatment plan clearly to the patient.
  • Discuss the importance of adherence to prescribed treatments and follow-up appointments.
  • Provide information on lifestyle changes that can improve health outcomes (e.g., smoking cessation, nutrition, exercise).

Conclusion

Evaluating a patient with symptoms of persistent cough, fatigue, and weight loss requires a systematic and thorough approach to ensure accurate diagnosis and effective treatment. By following a step-by-step method, healthcare providers can gather essential information, conduct necessary tests, and develop a comprehensive treatment plan tailored to the patient's needs. This methodical approach enhances the quality of care and improves patient outcomes.

Automating chain-of-thought prompting represents a significant advancement in the field of AI, enhancing the efficiency, accuracy, and usability of generative models. By integrating automated CoT frameworks, AI systems can handle complex queries more effectively, providing users with structured, reliable, and insightful responses. As this technology continues to evolve, it promises to transform various industries, making AI an even more powerful tool for problem-solving and decision-making.


The Power of “Take a Deep Breath and Work on This Problem Step-by-Step” in AI Prompting

Chain-of-thought prompting is a powerful technique that enhances the capabilities of generative AI models by guiding them through a structured reasoning process. By breaking down complex tasks into manageable steps, CoT prompting improves accuracy, understanding, and problem-solving abilities. This method has wide-ranging applications, from education and business to healthcare and beyond, making it a valuable tool for leveraging the full potential of AI in various domains.

One simple yet powerful sentence that can significantly enhance chain-of-thought prompting performance is: “Take a deep breath and work on this problem step-by-step.” Check why incorporating this sentence at the end of prompts is beneficial and how it can improve user interactions and AI.


Understanding the Importance of Calm and Methodical Processing

The phrase “Take a deep breath and work on this problem step-by-step” encourages a calm, methodical approach to problem-solving. Here's why this is important:

Promotes Clear Thinking:

By prompting the AI to metaphorically "take a deep breath," we emphasize the need for clarity and focus. This helps the AI avoid rushing through the problem and encourages a more deliberate and thoughtful response.

Encourages Sequential Reasoning:

Asking the AI to work step-by-step ensures that complex problems are broken down into manageable parts. This methodical approach aligns with the principles of Chain-Of-Thought prompting, where each step builds on the previous one, leading to more accurate and coherent responses.

Reduces Cognitive Load:

Complex tasks can be overwhelming. By instructing the AI to take it one step at a time, we reduce the cognitive load, making it easier for the AI to process information effectively and produce high-quality responses.


Practical Applications

Here are some practical examples where this sentence can be particularly beneficial:

Mathematics and Logic Problems

Prompt:

Solve the following algebra problem: What is the value of x in the equation 2x + 3 = 11? Take a deep breath and work on this problem step-by-step.

AI Response:

  1. Subtract 3 from both sides of the equation: 2x + 3 - 3 = 11 - 3.
  2. Simplify the equation: 2x = 8.
  3. Divide both sides by 2: x = 4.


Coding and Programming Tasks

Prompt:

Write a function in Python that checks if a string is a palindrome. Take a deep breath and work on this problem step-by-step.

AI Response:

  1. Define the function and accept a string as input.
  2. Remove any non-alphanumeric characters and convert the string to lowercase.
  3. Compare the cleaned string to its reverse.
  4. Return True if they are the same; otherwise, return False.

def is_palindrome(s): import re s = re.sub(r'[^A-Za-z0-9]', '', s).lower() return s == s[::-1]


Complex Decision-Making

Prompt:

Develop a project plan for launching a new product. Take a deep breath and work on this problem step-by-step.

AI Response:

  1. Define the project goals and objectives.
  2. Identify key milestones and deliverables.
  3. Determine the necessary resources and budget.
  4. Create a detailed timeline.
  5. Develop a risk management plan.
  6. Outline the execution strategy.


Medical Diagnosis

Prompt:

Evaluate a patient with symptoms of fatigue, weight loss, and persistent cough. Take a deep breath and work on this problem step-by-step.

AI Response:

  1. Collect detailed patient history and symptom information.
  2. Conduct a physical examination.
  3. Order relevant diagnostic tests (e.g., blood tests, chest X-ray).
  4. Analyze test results and correlate them with symptoms.
  5. Consider differential diagnoses.
  6. Formulate a treatment plan based on findings.


Psychological Impact on AI Interaction

Including calming phrases like “take a deep breath” can also have a psychological impact on users, promoting a sense of calm and encouraging a reflective approach. This can be particularly beneficial in high-stress scenarios, such as preparing for exams or dealing with complex work tasks.


Real-World Applications of Step-by-Step Reasoning

The step-by-step approach isn't just beneficial in theoretical contexts; it is also crucial in real-world scenarios. For instance, in legal practices, attorneys must present their arguments step-by-step to ensure logical coherence and persuasiveness. Similarly, in everyday tasks like solving complex riddles or managing projects, a step-by-step approach helps break down and systematically address each component.


Potential Pitfalls of Step-by-Step Reasoning

While step-by-step reasoning can be incredibly useful, it is not without its drawbacks. There are situations where it might lead to over-analysis or become counterproductive. For instance, individuals might lose sight of the bigger picture by getting too caught up in the minutiae. It's also possible that time constraints might not allow for a thorough step-by-step analysis, necessitating a more holistic and expedited approach.


Conclusion

Incorporating the sentence “Take a deep breath and work on this problem step-by-step” in AI prompts is a simple yet effective strategy to enhance AI performance. It promotes clear thinking, sequential reasoning, and a calm approach to problem-solving, leading to more accurate, comprehensive, and user-friendly responses. By integrating this technique, users can unlock deeper understanding and more effective interactions with generative AI, making it a valuable tool for various applications across education, work, and everyday problem-solving.

Need assistance with AI prompts and maximizing the potential of generative AI? Reach out to us at [email protected] for expert guidance and ready-to-use prompts.

#AI #GenerativeAI #PromptEngineering #ProblemSolving #ChainOfThought #Productivity #Innovation

Adrian Dionisio

Turn your Expertise & Knowledge into a Successful Business || I help Founders & Solopreneurs monetize their skill set || Consultant & Founder @business 737 || Dedicated to Business Owners World Wide

4 个月

By breaking tasks into smaller steps and processing them sequentially, it enhances accuracy and insight.

Chain-Of-Thought (CoT) prompting helps AI tackle tasks step by step, improving accuracy and insights. Give it a shot at MeclabsAI. Elizabeta Kuzevska

John Bush

B2B CEO, CRO Advisor | AI - Alignment | Sales & Marketing Automation | Revenue Transformation

4 个月

Thank you Elizabeta Kuzevska , as we get deeper into AI and its uses for our businesses the hard work is just beginning. Chain of thought prompting is a game-changer, making AI more reliable, transparent, and effective. We need more Experts like you leading the way.

Raj Vashisht

CEO at Adrianaa Services

4 个月

So insightful message thanks ???? very good morning ????

要查看或添加评论,请登录

社区洞察

其他会员也浏览了