AI Hints Are Poison (Unless You Use Them to Make Practice Harder)

You ask ChatGPT for a hint.

It gives you a neat nudge. You feel smart again. You move on.

And you just trained the wrong skill.

The dirty secret of JEE prep is that most students are not failing because they lack information. They are failing because they cannot retrieve the right idea under pressure. They cannot discriminate between similar-looking questions. They cannot choose the right tool fast enough.

A hint given too early does not fix that. It deletes the exact struggle your brain needs.

This is not anti-AI. Eklavya is literally built on AI.

This is anti-cheating-yourself.

If you use AI like a comfort blanket, it will make you feel productive while quietly stealing the thing you actually need: judgment.

What JEE actually tests (and why your practice lies)

JEE is not a memory test. It is not a handwriting test. It is not even mostly a “calculation” test.

It is a decision test.

• Which concept applies here?
• Which approximation is legal?
• Which assumption is hidden in this line?
• Which trap option is designed for your first impulse?

That decision step is what collapses under exam pressure.

In coaching culture, students are trained to worship speed. “Solve 100 questions.” “Finish DPP.” “Cover syllabus.”

Coverage is not competence.

Competence is being able to look at a new problem and pull the right move from memory, not because you saw it five minutes ago, but because your brain has built a retrieval route.

Cognitive science has been yelling this for decades. In a major review of 10 study techniques, John Dunlosky and colleagues rated practice testing and distributed practice as high-utility strategies because they reliably improve performance across tasks and learners. Highlighting and rereading, the comfort-food techniques, do not. (Dunlosky et al., 2013, PubMed PMID: 26173288)

So here’s the uncomfortable question:

When you ask AI for a hint, are you practicing retrieval, or are you outsourcing it?

Most students outsource.

The “help” that harms you

When you get stuck, you feel pain.

That pain is useful data. It tells you where your mental model breaks. It tells you what you cannot retrieve.

A well-timed hint can guide you. A poorly timed hint prevents learning.

Why?

Because the brain does not store knowledge like a folder. It stores routes.

Retrieval is the workout.

When you try to pull an idea out of memory, even if you fail, you are strengthening the pathway that makes future retrieval easier. That is the testing effect and retrieval practice literature in one sentence.

Karpicke and Blunt showed something that still annoys a lot of educators: students who repeatedly practiced retrieval learned more than students who used elaborative studying with concept mapping, even though concept mapping feels deeper and more “understanding-based.” (Science, 2011; PubMed PMID: 21252317)

If you replace retrieval effort with an AI hint, you are choosing elaboration without reconstruction. It feels like learning. It often is not.

That does not mean you should grind in silence for 45 minutes.

It means you should stop using AI as a first resort.

A blunt rule: delay the hint

Here is the rule I want you to follow for 30 days:

Do not ask AI for a hint until you have produced one full page of thinking.

One page.

Not vibes. Not “I kind of know.” Not staring.

Write:

• What the question is asking in your own words
• What chapter tags you think it is
• The first method that comes to mind
• Why it might fail
• A second method
• A third method
• The smallest sub-result you can compute right now

If you cannot produce a page, you do not have a “difficulty” problem. You have an avoidance problem.

And AI is enabling it.

Why a page matters

A page forces you to do two things coaching rarely trains:

1. Externalize your confusion so it becomes fixable.
2. Generate candidate moves so you practice discrimination.

Without candidates, there is nothing to discriminate.

With candidates, your brain is forced to choose.

That choice is the skill.

Use AI as an adversary, not a tutor

Most students use AI like this:

“Give me a hint.”

That prompt is too powerful. It invites the model to solve the decision step for you.

Instead, use AI to create desirable difficulty.

Your goal is to make practice harder in the right way, not easier in the comforting way.

Here are three modes that actually build exam skill.

Mode 1: Socratic gatekeeping

Prompt:

“Do not give the solution. Ask me one question at a time that forces me to choose the right concept. If I answer incorrectly, ask a sharper question. Only after 6 questions, summarize what I should have noticed.”

What this does:

• It delays the reveal.
• It turns help into retrieval prompts.
• It teaches you what cues matter.

In other words, it makes AI behave like a strict coach who refuses to rescue you prematurely.

Mode 2: Near-miss generator (discrimination training)

JEE punishes pattern-matching.

Two questions can look identical and require different ideas.

That is why students who rely on “I have seen this type” get wrecked.

Prompt:

“Given this problem, generate 5 near-miss variants that look similar but require a different concept or a different constraint. For each variant, do NOT solve it. Just state what changes and which concept becomes relevant.”

Now your brain is forced to answer: what is the pivot?

That is discrimination training.

Coaching rarely does it because it is hard to manufacture and even harder to grade.

AI can do it instantly.

Mode 3: Retrieval-only checks

Prompt:

“I will write a solution from memory. Your job is only to mark each step as (valid / invalid / unjustified). Do not fix it. If invalid, ask me what assumption I used and why it is wrong.”

This is uncomfortable.

Good.

It turns AI into a mirror, not a crutch.

The emotional part no one talks about

Students ask for hints early for one reason: fear.

Not fear of the exam.

Fear of feeling stupid.

Coaching culture trains you to treat confusion as failure. You are supposed to be a “serious student” who is always in control. If you are not in control, you panic and reach for help.

AI is the fastest help you have ever had. So you grab it.

Here is the truth:

Confusion is not a sign that you are behind. Confusion is the entrance to learning.

But only if you stay inside it long enough to build a route out.

This is where empathy matters.

If you are exhausted, if you are sleep-deprived, if you are carrying family pressure, your brain will choose comfort. Of course it will.

The answer is not self-hate.

The answer is a structure that makes the right behavior easy.

A 20-minute protocol for “getting unstuck” without lying to yourself

Use this during practice.

Step 1 (3 minutes): write your guess

Before anything else, write a guess.

Even if it is wrong.

Guess the chapter and the likely method.

This forces retrieval.

Step 2 (7 minutes): generate three candidates

Write three possible approaches.

• Approach A: the obvious one
• Approach B: the one you do not fully trust
• Approach C: the brute force or extreme case

You are not trying to be right. You are trying to create options.

Step 3 (5 minutes): run the smallest test

Pick the fastest test that can kill one approach.

• Plug a boundary value.
• Check units.
• Check symmetry.
• Check monotonicity.
• Sketch the graph.

This is the thinking coaching skips because it is not “syllabus.”

But this is what real problem solvers do.

Step 4 (3 minutes): ask AI one targeted question

Not a hint.

A targeted question.

Examples:

• “Which assumption is illegal in Approach A?”
• “What single observation would make this a conservation problem?”
• “What is the cue that distinguishes this from the similar-looking standard problem?”

If AI answers with a full solution, stop. Rewrite the prompt.

Step 5 (2 minutes): log the failure

Write one line:

“What cue did I miss?”

Not “I made a silly mistake.”

Not “I need to revise.”

A cue.

That cue is what you need to retrieve next time.

Why this works (and why it feels worse)

This protocol feels worse because it reduces fluency.

Fluency is the feeling of ease.

Fluency is also the most common liar in education.

Students feel fluent after rereading. They feel fluent after watching solutions. They feel fluent after AI hints.

Then they meet a new question and realize they cannot produce anything without the crutch.

Retrieval practice feels slower. It feels messier. It makes you confront gaps.

That is the price of real learning.

And it is the only price that JEE accepts.

One more uncomfortable point: your phone is not neutral

Even if you “use AI responsibly,” your phone is designed to fracture attention.

A lot of recent Eklavya posts have attacked this because it is destroying students quietly.

The research is mixed, too. A preregistered replication of Ward et al. (2017) found the famous “brain drain” smartphone proximity effect did not replicate in their setup (Ruiz Pardo et al., 2022; PubMed PMID: 36007374). Good. Science is supposed to correct itself.

But you do not need a perfect effect size to notice your lived reality: notifications, status anxiety, and constant novelty make deep practice harder.

So if you want to use AI without poisoning your prep, change the environment:

• Use AI on a laptop with everything else blocked.
• Or use AI in one scheduled window, not during every question.
• Or use AI on paper: copy the prompt into a notebook, then close the device and think.

The core idea is the same.

Protect the decision step.

If you remember one line

AI can make you faster at finishing questions.

It can also make you slower at doing them alone.

If you want a better rank, do not use AI to remove struggle.

Use it to design struggle.

Make practice harder in the right way.

Then walk into the exam with something coaching cannot sell you: real judgment.