There is a growing body of research examining how AI influences the way we think, learn, and solve problems.While no study has yet measured the long-term cognitive effects of AI use—after all, these tools have been widely available for less than five years—researchers are identifying patterns that align with established findings in psychology and neuroscience.I read through the research, identified the common mechanisms researchers think are behind these cognitive concerns, and built a handful of prompts to address them.
Instead of encouraging passive AI use, they're designed to keep you actively engaged in the thinking process—so AI helps strengthen your problem-solving skills rather than replacing them.I've linked every study I referenced during my research at the end of this article.Use it as a tutor—not an answering machine The difference between using AI and depending on it Close Instead of having AI write your code or essays, use it as a tutor that helps you become a better programmer or writer.
You can also use it to study the systems experts rely on and incorporate those approaches into your own work.If you're struggling with a particular concept, tell the AI exactly where you're stuck and ask for targeted guidance backed by research or established best practices.For example, I didn't feel comfortable letting Claude or ChatGPT vibe-code all my scripts and custom apps while I had no idea how they actually worked.
You wouldn't press Send on an AI-written email without reading it first, would you? Vibe coding can feel a lot like that—especially if you don't come from a programming background.So, instead of outsourcing the work entirely, I decided to learn how to code and turned Claude into my personal tutor.Now, learning is deeply personal, so a one-size-fits-all approach doesn't work.
I built my workflow around how I actually learn: by doing, not reading.I tell Claude what I want to learn, it creates a project-based curriculum, assigns me work, and acts as my mentor throughout the process.I complete the exercises, send screenshots of my progress, and ask for help whenever I get stuck.
Most importantly, I've configured it never to give me the answer outright.Instead, it nudges me in the right direction until I figure it out myself.Once I've mastered a concept, we move on to the next one.
I wrote up the entire workflow—including the full prompt—in my guide to turning Claude into the ultimate coding tutor, and it remains one of the most effective ways I've found to learn with AI.Related I transformed Claude into the ultimate coding tutor that tracks my progress (Prompt included) I built a coding tutor that won't let me cheat my way through it.Here's the prompt.
Posts By Dibakar Ghosh Deliberately track what you're offloading The audit nobody wants to run, but should Close Most LLMs, including ChatGPT, can now look back through your previous conversations.You can use them as an introspective tool by asking them to audit your usage and identify every instance where you've offloaded thinking.Suddenly, you have a map of the mental muscles you've stopped exercising—and you can make a conscious decision about which ones are worth keeping.
I personally believe that not all offloading is bad—and it depends on your priorities.I'm a writer, and writing for the web involves dozens of steps, from research to publishing.There are plenty of technical tasks—renaming images, adding tags, inserting internal links, formatting articles—that aren't particularly creative but still need to get done.
Those are tasks I'd happily delegate because I don't mind losing that particular muscle, especially when search engine best practices change so often that you're forced to relearn them anyway.The creative work is different.Finding topics readers actually care about, discovering the right angle for a story, structuring an argument, and crafting the prose are all skills I don't want to lose.
That's why I refuse to let AI do those parts for me—even when I’m staring at a blank document with a serious case of writer's block.Here's the prompt I use to track which tasks I'm gradually offloading to AI: Look back through my recent chats with you and identify every instance where I offloaded cognitive work to you: - Summarizing something I could have read - Writing something I could have written - Deciding something I could have reasoned through - Forming an opinion I could have formed myself Organize these into a table with three columns: the task, the underlying skill it exercises, and how often I've delegated it.Then flag which of these skills are likely to atrophy with continued offloading.
Don't soften the results to spare my feelings—I want an honest audit, not reassurance.End by asking me which of these skills I consciously want to keep exercising myself, so we can set rules for what I should stop delegating to you.Related I let Claude take over my PC—and it automated these 5 boring tasks for me Claude can do a lot more than answer your questions—it can automate the boring stuff, so you can focus on the fun stuff.
Posts By Dibakar Ghosh Challenge yourself with the opposite point of view Your AI is probably agreeing with you too much Close By default, most AI models lean sycophantic—and the more memory and context they build up about you, the better they become at telling you what you want to hear.Ask ChatGPT or Claude about a genuinely contested topic, and more often than not, it will respond with a perspective that closely aligns with your worldview.That can trap you in an echo chamber and limit your exposure to different perspectives.
It also shields you from testing your beliefs against strong opposing arguments, which is one of the best ways to sharpen your thinking.Sometimes you'll win and come away with stronger convictions.Other times you'll realize your position doesn't hold up under scrutiny.
Either outcome is intellectual growth, but a sycophantic answering machine can shield you from that process.Fortunately, this is a much simpler problem to solve.When you're exploring an idea, instead of asking the AI what it thinks, tell it what you think and assign it the opposite side of the debate.
Since chatbots are excellent at roleplaying, they can take on the perspective of someone who genuinely holds the opposing view and give you a civil debate focused purely on the ideas—without the name-calling or personal attacks that often derail online discussions.In fact, I think this is one of AI's best use cases.Most of us don't get this kind of experience in our everyday lives because we avoid difficult conversations with friends and family.
There are also times when we stay quiet because we don't feel informed enough to contribute, forgetting that everyone starts out knowing less than they eventually will.With an AI, those social pressures largely disappear, making it much easier to ask naïve questions, challenge your assumptions, and refine your thinking.Here's the prompt I use when debating with Claude and ChatGPT: I want to explore a topic through structured debate.
My position is: [state your position].Your job is to argue the strongest possible case for the opposing view—steelman it, don't strawman it.Use the best arguments, evidence, and reasoning that actual proponents of that side would use.
Stay civil and professional at all times, but do not soften your arguments to agree with me, and do not concede points just because I push back.If my rebuttals have genuine weaknesses, point them out directly.After every few exchanges, briefly summarize where each side stands and which arguments remain unresolved.
Do not break character to validate my position at any point.Create a Claude Project, a ChatGPT Custom GPT, or a Gemini Gem called "Debater" and save this prompt as its custom instructions.Don’t paste this prompt in regular conversations or else the AI can build up long-term memory that you prefer this behavior and start challenging you during everyday conversations when you're looking for collaboration instead.
Don't let it fish for you—use it to learn how to fish for yourself The point I want to drive home is that AI is neither good nor bad—it's a tool, and an incredibly powerful one at that.If you use it carelessly, it can slowly erode your skills.If you use it deliberately, however, it can help you build new ones.
Studies I referenced in my research Generative AI without guardrails can harm learning: Evidence from high school mathematics Endoscopist deskilling risk after exposure to artificial intelligence in colonoscopy: a multicentre, observational study Experimental evidence of the effects of large language models versus web search on depth of learning Effects of LLM use and note-taking on reading comprehension and memory: A randomised experiment in secondary schools Cognitive ease at a cost: LLMs reduce mental effort but compromise depth in student scientific inquiry The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers Generative AI enhances individual creativity but reduces the collective diversity of novel content The effects of over-reliance on AI dialogue systems on students' cognitive abilities: a systematic review AI Tools in Society: Impacts on Cognitive Offloading and the Future of Critical Thinking
Read More
