You got into science to think. To question. To sit with a problem until the world fell away. But your AI assistant is making sure that never happens.
Here’s the silent pact you didn’t sign: you hand over the boring, repetitive work to an LLM — and in return, you’re supposed to get more time for the deep, creative stuff. That’s the promise. The tool meant to free your mind is quietly filling it with noise.
A recent study from the authors of arXiv:2607.17397 confirms what many of us have felt but couldn’t name: LLMs increase raw productivity — you can write more code, draft more emails, plow through more literature — but the time you save doesn’t become thinking time. It becomes more work. More requests. More meetings. More output.
I watched it happen to a friend. She’s a computational biologist. Started using LLMs to automate data cleaning, code generation, paper formatting. Saved three hours a day. Those three hours were immediately consumed by reviewers asking for more analyses, colleagues asking for help, and the endless treadmill of ‘productivity.’ She ended up publishing more papers, but she can’t remember the last time she spent an afternoon just staring at a whiteboard.
This is the efficiency trap. You optimize yourself into a machine that produces, but never creates. The very thing that was supposed to amplify your intellect quietly erodes your ability to think deeply.
Most productivity gurus will tell you to automate everything. They’re wrong. Neutrality is death here. I’m taking a side: the relentless pursuit of efficiency is making science faster, but it’s making scientists shallower. The breakthroughs that change fields don’t come from optimizing your workflow — they come from the slow, unstructured, almost wasteful moments of reflection that we are systematically eliminating.
Consider the paradox: we have more tools than ever to help us think, yet we have less time to actually think. Every saved minute is immediately filled with a new demand. The LLM doesn’t give you freedom — it gives you a faster leash. Efficiency is the enemy of insight.
What’s the alternative? Deliberate inefficiency. Schedule time where you explicitly turn off the LLM. Sit with a blank page. Stare out a window. Let your mind wander. The next breakthrough won’t come from a faster workflow — it will come from the quiet moment you didn’t fill.
I’m not saying ditch the AI. I’m saying stop pretending it’s making you smarter. It’s making you faster — and there’s a difference. Your most valuable resource isn’t speed. It’s the capacity to be slow. Protect it like your career depends on it, because it does.
FAQ
Q: Does this mean I should stop using LLMs entirely?
A: No. Use them strategically for tasks that truly don't require deep thought. But deliberately protect unstructured time — schedule 'slow hours' where you turn off all AI tools and let your mind wander. The goal is balance, not abstinence.
Q: How can I measure if my LLM usage is hurting my creativity?
A: Track your 'deep work' hours — uninterrupted time spent on complex, novel problems. If they're decreasing even as your output increases, you're in the trap. Also, ask yourself: 'When was the last time I had a truly original idea that didn't come from iterating on something the AI generated?'
Q: Isn't it just a matter of self-discipline? If I save time, I can choose to think. The tool isn't forcing me to do more work.
A: In theory, yes. In practice, the social and institutional pressures to be 'productive' (more papers, more grants, more code) fill the vacuum. The LLM amplifies the system's demand for throughput. You need to actively resist that pressure — it won't happen organically.