How to Avoid AI Brain Rot - Miles Cranmer

Summary

Miles Cranmer argues that AI can improve the work people produce while weakening their ability to do or assess that work themselves. Because many people now have access to similar tools, using AI in the default way can also make work more alike and pull performance toward the average. The goal, he says, is to use AI to extend your abilities—not replace your own thinking.

Learning and self-development strategies

• Do the thinking before asking for the answer. Struggling with a problem helps build understanding and long-term recall, even when it feels less productive.
• Use AI as a tutor, not an answer machine. Ask it to quiz you, give one question at a time, and offer hints only when you are stuck. Explain your reasoning back to it, then revisit the material and try again.
• Protect skills you are meant to learn. If a task is part of your learning, attempt it yourself first. For other tasks, delegate only when you can check the result.
• Use the “explain and defend” test. Before relying on AI output, ask whether you could explain it, defend it, and identify likely mistakes without AI.
• Turn AI into a visual learning aid. Ask it to create plots or small interactive demos to help explore a concept, but verify that they are correct.
• Treat AI as one learning tool among several. It can make mistakes, especially with specialized material, so check its explanations against reliable sources.

Working with AI and coding agents

• Use AI to accelerate your work, not substitute for your judgment. Humans remain responsible for AI-generated output.
• Set clear standards in your instructions. For learning, ask the AI to quiz you rather than give solutions. For coding, specify expectations for the work and its review.
• Inspect results in a form you can evaluate quickly. Plots can reveal faulty data-analysis results; flowcharts may help trace program logic. Choose the check that fits the task.
• Understand a bug before fixing it. Reproduce it and identify its underlying cause rather than letting an agent patch only the symptom.
• Ask coding agents for the least code that works. Cranmer cautions that agents may produce overly complicated solutions that pass tests without being well designed.
• Keep human oversight of codebases. Cranmer says agents can introduce technical debt and struggle to maintain long-term architecture without people reviewing their work.

Adaptation and ambition

• Keep learning and adapting. Since AI tools change quickly, being able to pick up new skills may matter more than mastering only today’s tools.
• Use AI to tackle harder problems. Rather than focusing only on automation, Cranmer encourages using the time and capabilities AI provides to pursue more ambitious scientific and other challenges.

Cranmer points to studies suggesting that AI can improve performance on practice tasks while harming later exam performance when students rely on solutions. In contrast, hint-based tutoring and asking for explanations were associated with better learning outcomes. He also describes findings that AI can raise average output while making work more similar, and that experts’ judgment helps them use AI more effectively.

Presenters and sources

• Presenter: Miles Cranmer, assistant professor of science.
• Sources discussed: Several studies on AI use, learning, memory, writing, and productivity were described, but full citations were not provided in the subtitles.
• AI tools and workflows mentioned: ChatGPT, Claude, Codex, Ponytail, and P Stack/Pystock (the last name is unclear in the subtitles).