The Cost Of UPSC Prep - UPSC Killed My Youth, Wasted Years & Life After UPSC

ELI5 / TLDR

India’s most prestigious government-job exam, the UPSC civil services test, just held its 2026 first-round paper, and a teacher who himself cleared the early rounds six times says it has quietly broken. The questions, he argues, no longer connect to any syllabus or any book a student could have studied — so the people who pass are increasingly the lucky guessers, not the hard workers. The conversation then widens into how this leaves millions of young Indians blaming themselves for failing an unfair test, and drifts into a frank exchange about free speech, government control, and why nobody in power seems to be listening.

The Full Story

The guest is Amit Kilhor, a UPSC coaching teacher who says he personally cleared the preliminary stage of the exam six times. The host is Ravi Kapoor, a serving Indian Revenue Service officer who runs the channel. The framing is personal — Kilhor says he started his own Instagram after watching Kapoor’s videos — but the meat is a forensic takedown of the exam that decides who becomes a senior Indian bureaucrat.

The exam is no longer connected to anything you can study

Kilhor’s central claim is simple. A good hard exam is one where, when you fail, you know why you failed — you can point to the chapter you skipped. The 2026 Prelims, he says, is not that. It is an exam where you fail and have no idea why, because the questions trace back to no syllabus and no book.

“A tough question is one where I know why I couldn’t do it. This is different — here you don’t even know the question.”

His examples are concrete. One question asked which country a Nobel laureate was born in — not what field he won in, not what he invented, just his birthplace, a fact buried in a single line of one newspaper article all year. Another asked about UN peacekeeping missions from 2007 to 2010. The complaint isn’t that the questions are hard; it’s that they’re random. A student who read deeply for two years gets no advantage over one who didn’t.

He frames the syllabus problem bluntly. For fifteen years, he says, UPSC has described its Prelims syllabus in roughly seven vague words — “current affairs,” “history,” “geography,” and so on — without naming a single reference book. So a student is told to “know history” but the questions don’t come from any standard history textbook. Imagine being told to study for a physics exam and then being handed a medical paper.

The cut-off is the tell

There’s a clean diagnostic buried here. The “cut-off” is the minimum score needed to pass. Kilhor’s argument: when the cut-off is high (say 116 out of 200), it means the paper was aligned to the syllabus — students who studied could score, so the bar rose. When the cut-off collapses to 70 or 75, it means even the well-prepared could only solve about 40 of 100 questions. A low cut-off, in his reading, is not a sign of a “tough” paper. It’s a sign of a broken one — proof that effort stopped mapping to marks and luck took over.

“If the cut-off is 75, it means the syllabus had no relationship to the paper.”

The reliability problem

He makes a sharper, almost academic point about what a fair exam even is. An exam is reliable only if it repeats — the same student should score roughly the same if they sat it again. His damning example: candidates who were selected last year, who are days away from starting their IPS (police service) training, fail this year’s Prelims. If the same people pass one year and fail the next on equivalent papers, the test isn’t measuring ability — it’s a lottery.

Who gets hurt — and the Hindi-medium problem

A recurring theme is unfairness to Hindi-medium students. Kilhor claims the papers are written in English and then translated, often badly, every year — so the Hindi version is effectively broken, and the books a Hindi student reads were never the source of any question to begin with. He cites how few Hindi-medium candidates clear the exam (a handful a year out of thousands) as evidence of a structural, unannounced bias. He’s careful here: he says good discrimination in an exam (separating strong from weak on clear criteria) is fine and necessary, but this isn’t that — there’s neither clear criteria nor real differentiation, just noise.

He also flags small but telling sloppiness: the official 2026 notification, he says, had over a hundred spelling mistakes when he ran it through a checker. Not a crime in itself, but coming from the body that decides careers, it signals an organization that isn’t sure of what it’s doing.

Why does UPSC behave this way?

The diagnosis is institutional, not conspiratorial. Kilhor thinks UPSC was once a genuinely student-friendly, responsive body, and that something broke in the last five or six years. His read: an overworked institution that’s tried to “innovate” and make the exam “more aligned with reality,” didn’t know how, and went sideways. He suspects they prize internal secrecy so heavily — to avoid paper leaks and litigation — that their own quality control has decayed. They drop a dozen-plus flawed questions every paper after the fact rather than fixing them before printing.

There’s a striking aside on AI. Kilhor says he fed one contentious question (which ancient text a term first appeared in) to multiple large language models and got different answers — ChatGPT said Rigveda, Google said Atharvaveda, Claude said something else. His point: if the billion-dollar models disagree on the “right” answer, what hope does a student have, and how confident can UPSC be? He even speculates UPSC may be using some AI-driven option-randomization that’s injecting subtle errors.

The pivot to free speech

The second half drifts away from the exam into politics. Kapoor pushes Kilhor on why he speaks out, and whether the recent online UPSC outrage is organic or a planted political movement (he floats the Nepal-style youth uprising and opposition parties as possibilities). Kilhor’s answer: it has to be organic, because you can’t buy twenty million followers — if you could, the ruling party itself wouldn’t have fewer.

That opens a candid exchange on Indian free speech. Kilhor rates it around 5.5 or 6 out of 10 — better than nothing, far below the US or Europe. His model of government behavior is a “pressure cooker”: the state allows criticism up to a point because letting steam out is safer than clamping down, which pushes dissent underground (he cites Nepal, where shutting down social media backfired). When a critical voice crosses a tipping point and threatens “internal security,” he believes the state quietly breaks the momentum — not through hard suppression, but by tracking and containing.

He closes the political thread with a clean, non-partisan jab on fuel prices: the government cut petrol prices right before an election, but sat on cheap oil for three prior years and waited two months through the election cycle to act. His stance throughout is that he criticizes whoever deserves it, loves the judiciary, and feels a moral obligation to speak for the 9,999 students who fail silently for every one who passes and gets a microphone.

“In the UPSC exam, 10,000 kids sit, one becomes an officer, and everybody listens to his story. Nobody is ready to hear the other 9,999.”

Key Takeaways

A fair hard exam tells you why you failed (a chapter you skipped); an unfair one leaves you unable to locate your mistake. The 2026 UPSC Prelims is, per the guest, the second kind.
UPSC’s Prelims “syllabus” is roughly seven vague words with no named reference books — so students can study for two years and still face questions sourced from nowhere in their material.
The cut-off score is a diagnostic: a high cut-off implies syllabus-aligned questions; a collapsed cut-off (70–75) implies the paper had no real relationship to what students studied.
Test reliability means repeatability — the same candidate should score similarly on equivalent papers. Candidates selected last year failing this year’s Prelims is cited as proof the test has become a lottery.
Hindi-medium candidates are structurally disadvantaged because papers are written in English then poorly translated, and questions trace to English-medium sources.
UPSC drops 12+ flawed questions per paper after administration; the guest argues that proves they could catch errors before printing but don’t.
Large language models disagree with each other on some “factual” exam answers — used as evidence that the questions themselves are ambiguous, not that students are underprepared.
Proposed fixes: align questions strictly to a published syllabus, name reference books, and move to computer-based testing (feasible even at ~800,000 candidates).
The “pressure cooker” model of state control: allow criticism as a release valve up to a tipping point, then quietly contain rather than openly suppress, because hard suppression drives dissent underground (Nepal as the cautionary case).
The guest rates Indian free speech ~5.5–6/10 — functional but heavily below developed democracies and skewed by corporate-controlled legacy media.

Claude’s Take

The exam critique is the strong half, and it’s genuinely good. Kilhor isn’t ranting — he reaches for real testing concepts (reliability, repeatability, criterion validity) and lands the cut-off-as-diagnostic argument cleanly. The “candidates who passed last year fail this year” point is the kind of falsifiable, structural evidence that’s hard to wave away. If even half his claims about syllabus opacity and translation rot are accurate, the indictment stands.

The weaknesses are the usual ones for an aggrieved insider. He’s a coaching teacher, which is a stakeholder, not a neutral party — though to his credit he defends coaching institutes as symptoms rather than villains. Everything is asserted from memory and anecdote; there are no linked papers, datasets, or side-by-side question analyses, so you’re trusting his recall of specific questions and cut-offs. And the heavy emphasis on Hindi-medium bias, while plausible, is argued from outcomes (few selections) rather than a controlled look at the translations themselves.

The free-speech second half is more vibes than rigor. The pressure-cooker model is a tidy intuition, but it’s just-so storytelling — unfalsifiable, and the same observations would fit several other explanations. The oil-price jab is the most concrete and the most clearly partisan-adjacent, even as he insists he’s non-partisan.

Score 6. The exam analysis alone is worth the time for anyone curious about how a high-stakes Indian institution decays from the inside, and the cut-off and reliability framings are portable to thinking about any selection test. It loses points for being one-sided by construction, unsourced, and for a back half that wanders into speculative politics. Also worth flagging: this was a Hindi conversation transcribed by AI, so specific numbers (cut-off figures, question details, the spelling-mistake count) should be treated as approximate.

The Cost Of UPSC Prep - UPSC Killed My Youth, Wasted Years & Life After UPSC | Ft. Amit Kilhor