Using ChatGPT As a Copilot For Your Mind

Episode Summary

Executive Summary: In this episode of The Cognitive Revolution, host Nathan LeBenz interviews Dan Shipper, CEO of Every, about practical uses of ChatGPT. They discuss AI as a co-pilot versus delegation mode, using ChatGPT for writing, coding, and research. Dan shares techniques like using AI for micro-tasks, summarizing, and creating diagrams. They also explore the future of AI, including state-space models and the potential for radical access to expertise.

Main Topics: AI as Co-pilot vs. Delegation Mode (Priority: 5/5): Discussion of two primary modes of AI interaction: co-pilot mode (real-time assistance) and delegation mode (offloading tasks entirely). The gap between them is expected to close with better agents. Using ChatGPT for Writing and Creativity (Priority: 5/5): Dan Shipper explains how he uses ChatGPT for micro-tasks like summarizing ideas, creating outlines, and finding metaphors. He emphasizes that AI helps with drudgery but requires human oversight for final output. Practical Coding Assistance with ChatGPT (Priority: 4/5): Nathan shares his experience using ChatGPT to build a React app module (a prompt coach) despite having no prior React experience. The AI helped with file structure, code generation, and debugging, saving significant time. Creating Diagrams for Patent Applications (Priority: 3/5): Nathan used ChatGPT to generate graphviz syntax for diagrams representing his company's AI workflow. This process helped clarify the system architecture and was used for a provisional patent application. Research and Fact-Checking with Perplexity (Priority: 3/5): Nathan discusses using Perplexity AI for accurate, fact-based queries, such as researching used minivan features. He contrasts it with ChatGPT's browsing capabilities, noting Perplexity's speed and accuracy. Future of AI: State-Space Models and Expertise Access (Priority: 4/5): Nathan predicts that AI will provide radical access to expertise (e.g., medical diagnosis) and that new architectures like state-space models (Mamba) will complement transformers, leading to further progress.

Key Arguments: AI can offload cognitive drudgery in knowledge work, similar to how machines offloaded physical labor in the Industrial Revolution. Using ChatGPT for micro-tasks (e.g., summarizing, outlining) is more effective than expecting it to produce final output wholesale. Prompting best practices include allowing the model to reason step-by-step (chain-of-thought) rather than forcing immediate answers. Custom instructions in ChatGPT can significantly improve relevance by providing context about the user's identity, goals, and weaknesses. Perplexity AI is superior to ChatGPT for fact-based queries due to its speed and accuracy, making it a practical tool for research. The future of AI includes composite architectures (e.g., transformers + state-space models) and radical access to expertise, which will be transformative but also disruptive.

Data Points: Time saved in coding task: 80-90% - Nathan estimated that building a React app module with ChatGPT took 2-3 hours versus 2-3 days without AI. AI vs. human diagnostic accuracy: 60% vs. 30% - In a Google DeepMind study on differential diagnosis from medical journals, AI outperformed human clinicians by a significant margin. AI vs. human radiologist performance: 60% to 40% - Human radiologists beat AI radiologists in reading x-rays and tissue slides, but the margin was narrow.

Pivotal Quotes: "I think one of the things it's really good at is pointing out the obvious solutions that you missed because you're too close to the problem." — Dan Shipper: Dan explains how ChatGPT can help generate basic outlines that reveal simple structures overlooked due to overthinking. "The goal of delegation mode is to get the output to the point where it is consistent enough that you don't have to review every single output." — Nathan LeBenz: Nathan describes the ideal state for delegating tasks to AI, where human oversight is minimized. "We are apparently headed for a world where you should be able to access that AI doctor. And if it's a 2x better performance on such a challenging task as differential diagnosis, that I think we're headed for a world of radical access to expertise." — Nathan LeBenz: Nathan discusses the potential for AI to democratize access to high-quality medical expertise.

Implications: Listeners should explore using AI for micro-tasks in their workflows, adopt chain-of-thought prompting, and consider tools like Perplexity for research. The future promises radical access to expertise and new AI architectures, but society must prepare for disruption in labor and equality.