The CORRECT way to use ChatGPT (in 2025)

Summary

In this video, Jeff simplifies the process of choosing between ChatGPT's various models and features. He explains when to use basic chat models versus reasoning models, offering real-world examples to guide viewers. The video covers essential tips for using reasoning models effectively, how to leverage the web search feature for context, and the power of deep research and canvas features for more complex tasks. Additionally, Jeff highlights tools for improving text output and offers advice on selecting the best AI tools for different needs, including Google Gemini.

Takeaways

• 😀 Reasoning models are best for complex, high-stakes queries, while basic chat models are suitable for quick, low-stakes tasks.
• 😀 Most users should default to using reasoning models, even if it means waiting slightly longer for better results.
• 😀 When prompting reasoning models, avoid telling them to think step-by-step, as they already do so inherently.
• 😀 Delimiters are helpful in separating instructions and content when using reasoning models, improving clarity and accuracy.
• 😀 Examples are optional for reasoning models, only use them when you're getting wrong results.
• 😀 ChatGPT’s web search feature is great for fact-based queries with explanations, but Google is still faster for basic facts.
• 😀 When researching, use ChatGPT’s deep research feature to analyze and summarize complex topics from multiple sources.
• 😀 The canvas feature is ideal for iterative work, such as refining documents, preparing reports, or editing content multiple times.
• 😀 Using built-in shortcuts in the canvas feature can help make quick changes and improve your workflow efficiently.
• 😀 For the best results with deep research, provide detailed prompts, or use external sources like PDFs for guidance.
• 😀 Utilize commands like 'elaborate', 'critique', and 'rewrite' in text-to-text models to refine content or check for flaws.

Q & A

Why should most people default to the latest O-number reasoning model with the cleanest name?

The latest O-number reasoning models tend to perform the best due to their refined and optimized capabilities. They are designed to handle complex queries more effectively, providing better responses with less 'baggage' than older or less updated models.

When should you use a basic chat model instead of a reasoning model?

A basic chat model is ideal for low-stakes tasks where you need a quick response, like simple factual queries. It's best when the accuracy of the answer isn't critical, and you don't mind slight deviations in the response.

What kind of queries benefit from using a reasoning model?

Reasoning models are better for tasks that require detailed thought, nuance, or more complex analysis, such as creating tailored advice, handling intricate queries, or providing insights that demand understanding and reasoning.

What is the significance of using delimiters when prompting reasoning models?

Delimiters help separate instructions from the content that needs to be analyzed, making it easier for the reasoning model to distinguish between what it should do and what it should analyze. This improves the quality of responses.

Why should you avoid saying 'think step by step' when prompting reasoning models?

Reasoning models are inherently designed to think step by step without being explicitly told to do so. Including this phrase can unnecessarily complicate the prompt and decrease the model's performance.

What does the 'zero-shot' approach mean in the context of reasoning models?

'Zero-shot' refers to prompting a reasoning model without providing any examples. Reasoning models are capable of handling such prompts effectively without needing prior examples, unlike chat models that may require examples for better performance.

When is it better to use ChatGPT's web search feature rather than Google Search?

ChatGPT's web search is useful when you need a fact with an explanation or context. Google is faster for quick, straightforward facts, while ChatGPT excels in providing additional context or insights around a specific query.

What is the ideal use case for ChatGPT's deep research feature?

Deep research is most effective when you need an in-depth report or analysis on a specific topic, such as comparing earnings reports from multiple companies or analyzing complex data. It pulls from various sources to create detailed, comprehensive answers.

What is the Canvas feature in ChatGPT, and when should it be used?

Canvas is a feature that allows for iterative editing and building upon responses. It's ideal for situations where you expect to make multiple revisions, such as drafting a document or preparing for a performance review.

What are some effective pro tips for using the Canvas feature?

Pro tips include using the back and forward buttons to navigate between versions, utilizing built-in shortcuts to make changes, and downloading the final document in markdown format for easy formatting and use in other applications.