Learn by Doing
I shared a prompt with my students at the Gokhale Institute the other day, and I’m sharing it here with you guys as well. The idea was to build an AI tutor who would be able to help the students on an ongoing basis with their studies in statistics. Only partly tongue in cheek, I named the file “Put Kulkarni Out of a Job”.
Please, go over the prompt, experiment with it, tweak it, and share it with folks who might find it useful. If you think you’ve made it better, please do let me (and others) know how, so that we can all learn better.
Once you play around with it, you will realize that you can (and should) customize it to suit your hobbies, learning styles and conversational preferences. Even better, you will realize that you can do this for all subjects that you’re studying right now.
Best of all, you will realize that nothing prevents you from choosing to learn any subject under the sun. Use a modified version of this prompt, upload the textbook of your choice in order to get RAG-based answers, and go ahead and learn whatever you want, at your own pace, whenever you want. The prompt (and more importantly, the LLM, so long as it is any paid version) is already good enough to teach you any subject you like.
Your Own Personal Collection of PhDs
As Aidan puts it:
demis is ofc right here: no one should fire their phds for gpt-5
rather, we’ve democratized the experience of being a 10 year old growing up in cambridge who can ring random doorbells and ask experts about hologram theory or sumerian history
everyone gets a phd in their pocket https://t.co/lqmRjhvFGk— Aidan McLaughlin (@aidan_mclau) September 14, 2025
Not only does everyone get a PhD in their pocket, but you can train your PhD to answer questions in your preferred style. Your personalized collection of bespoke PhDs will sit in your pocket, and will wake up at a moment’s notice, 24/7, to answer any question you may have. They will do this with text, audio and soon, video (and who knows what else, in very short order).
This generation of students, in other words, is cursed and blessed at the same time. They can no longer say that they had a bad prof for a particular subject, because they can always build a better prof for themselves. You can have an opinion about whether the prof that has been built is better or not along all dimensions, and if you want to shake your fist at the sky about it never being as good as a human prof, go right ahead.
But here are the facts: at INR 2,360/- per month, you will not get any human prof you can wake up anytime you want, who will answer however many questions you have with infinite patience, infinite variations and infinite applications, all based on your textbook of choice. These AI profs will check your work, design additional questions, conduct vivas for you, and they will do this without losing patience.
As a prof, I need to help my students understand that this is not just possible to do for all the subjects they’re learning, but that they’ll be missing out if they do not do this.
Create infinite and personalized office hours for yourself
Solve a problem from the textbook, take a photo, and upload your solution to have it checked by the prof you’ve built.
It’ll check your work, give you step-by-step confirmations (or corrections, if applicable), and ask some quick follow-up questions to make sure you’ve really understood the concept.
It’ll tell you what you can read next, and give you a set of flashcards related to the problem you just solved. Again, if any of this is not to your taste, changing it is trivial.
If typing is not your thing, you can chat with it using voice instead. You can upload more than one textbook, if you like. You can ask it to help you learn Python by solving the same problem using Python code:
Python’s not your thing, and your college requires you to learn Excel or SPSS instead? Sure!
Or all three, if you prefer. Or Stata. Or something else.
LFE, So Far
Again, once you learn how to learn with AI, you can learn almost anything, and you no longer need depend on the subject you want to learn being offered by the college you are enrolled in.
And every single one of you can do this, starting right now, for INR 2,360/- per month.
There are going to be a lot of people who will be working on making this process easier. The process, that is, of abstracting away the scaffolding needed to make learning as seamless as possible. These models themselves will get better and better over time, but adding in features that will help (especially if you have prior experience of what the typical pain points of students are like) is now ridiculously easy.
Hell, that’s exactly what LFE is!
Doing Matters
But the proof of the pudding is in eating it. And similarly, the proof that you have learnt something lies in being able to build something with what you know.
If learning becomes easier than it was in the past…
…does that then mean that the new bottleneck is in showing that you have learnt?
How do you show that you have learnt?
Ansh and I have some answers to this question, and in tomorrow’s post, we hope to show you what we’re trying to do.