I’ve been using ChatGPT-5 wrong. It’s actually insane (full guide)

About this video

• Video Title: I’ve been using ChatGPT-5 wrong. It’s actually insane (full guide)
• Channel: Alex Finn
• Speakers: Alex Finn
• Duration: 21:53

Overview

This video explains how to effectively prompt GPT-5 by highlighting its differences from previous models. The presenter introduces four key prompting techniques: controlling reasoning level, verbosity, tool calling, and self-reflection (using a rubric), along with a fifth technique of metaprompting. A free tool for optimizing prompts is also showcased, promising significantly improved results when these methods are applied.

Key takeaways

• GPT-5 is a router model: Unlike previous versions that offered different modes for speed or reasoning, GPT-5 is a single, powerful model that requires explicit control over variables like reasoning level and verbosity.
• Control Reasoning and Verbosity: Users must specify the desired reasoning depth and length of the response (low, medium, high verbosity) in their prompts, as GPT-5 does not auto-determine these.
• Leverage Tool Calling: GPT-5 can utilize multiple tools (e.g., image generation, web search, PDF creation) within a single prompt, a capability that should be exploited for complex tasks.
• Implement Self-Reflection (Rubrics): Instructing GPT-5 to create and use a self-grading rubric before generating a final response can lead to significantly better and more refined outputs, especially for code generation.
• Utilize Metaprompting: If initial prompts yield unsatisfactory results, use metaprompting to ask GPT-5 to revise its own prompts by specifying desired versus actual outcomes.
• Use the Prompt Optimizer: OpenAI offers a prompt optimizer tool that refines prompts specifically for GPT-5, enhancing results without manual prompt engineering.