JEE, CGPA, GATE, Codeforces: india’s rank fetish is killing technical talent

i don’t mind working with people who don’t know things. nobody knows most things. you can teach someone a protocol, send them a paper, point them at source code or spend an afternoon inside gdb figuring something out together. what is painful is working with people who don’t know, don’t put in the effort to know, and have somehow convinced themselves they’re exceptional.

india is extremely good at manufacturing this because we don’t merely use examinations, we have an examination fetish. rank is the fetish object. AIR, CGPA, GATE score, codeforces rating, package. give the system a number it can sort people by and everyone gets strangely excited.

JEE is probably the funniest example. the normal route into computer science at an IIT begins with an examination of physics, chemistry and mathematics. computer science is not even on the fucking paper. you could spend your teenage years writing compilers, finding vulnerabilities, contributing to kernels, reversing binaries or building distributed systems and none of that directly exists to the filter, but solve enough rotational mechanics quickly enough and apparently we have discovered your potential as a computer scientist¹.

JEE is difficult, obviously. doing well at it demonstrates mathematical ability, discipline and performance under a very particular kind of pressure. india also needs some scalable way of allocating scarce seats. the fetish begins when the proxy becomes the person. you cleared JEE, therefore you are intelligent. you got CSE, therefore you understand computers. you got IIT, therefore somebody outside IIT probably has less to teach you. we took a sorting algorithm and gave it spiritual significance.

then everyone acts surprised when the computer refuses to participate in the kink.

i’ve written about this before. the api does not care about your phd. gcc doesn’t know your AIR. linux doesn’t ask your CGPA before exposing the scheduler source. if a seventeen-year-old understands a subsystem better than you, then for that subsystem the seventeen-year-old is better than you. the machine has no respect for seniority.

sagnik pointed out another consequence i really like: we take the coaching-centre worldview that got us into college and reproduce it inside programming communities. which dsa sheet? when should i start competitive programming? what codeforces rating gets shortlisted? which project gets internships? is distributed systems “needed” in first year? there is always another roadmap, another bhaiya, another checkpoint where somebody slightly older gives you permission to learn the next thing.

competitive programming isn’t the problem. leetcode isn’t the problem. wanting money isn’t the problem. needing somebody else to give you a fucking syllabus for curiosity is.

computer science has no age-locked skill tree. you don’t unlock distributed systems in third year. /proc doesn’t check your semester. redis doesn’t demand your transcript before letting you read the source. cloudflare does not ask your JEE percentile before dropping your packets. learning is asynchronous, and the internet made the hierarchy even more ridiculous because the same source code, papers, lectures and documentation are sitting there for everyone.

when everything becomes a rank, people start fucking the metric instead of getting good at the underlying thing. codeforces and codechef communities have repeatedly complained about leaked solutions, telegram groups and people buying live answers for rating; one indian competitive programmer explicitly connected some of this to resume shortlisting². there have also been public accusations involving students from elite indian institutes using leaked solutions³.

JEE obviously didn’t cause codeforces cheating. the pathology is broader. we trained people to fetishize the scoreboard: AIR, CGPA, codeforces rating, leetcode count, package. if the number gets rewarded, somebody eventually optimizes the number harder than the capability it was supposed to represent. goodhart would have a fucking aneurysm.

cybersecurity makes this even more absurd. CERT-In’s 2026 Scientist-B recruitment shortlisted candidates using GATE scores, and the final merit calculation gave 60% weight to GATE and 40% to the interview⁴. GATE CS tests useful computer-science fundamentals, but it does not tell me whether somebody can reverse a binary, fuzz a parser, notice a broken trust boundary, reason through oauth, understand a pcap, write a detection or stare at one fucked-up HTTP response until 3 am because something feels wrong.

an exceptional hacker can be rejected before anyone asks them to hack anything. bureaucracy loves competence only after competence has filled the correct form.

this is why IIT Kanpur’s B.Cyber experiment is genuinely interesting to me. before B.Cyber, i remember seeing the same exam-shaped security mindset even around IITK: one-shot SQL injection questions, people asking what exact material they should prepare, security getting reduced into another chapter to finish before the test. B.Cyber at least tries to break some of that. the admission process considers previous cybersecurity experience and then puts candidates through a six-hour hackathon involving cryptography, web/API/network security, reverse engineering and osint⁵.

show me the CTFs. show me the research. show me what you built. show me what you broke. open the shell, hand them the binary and see what happens. make the hacker touch the fucking computer.

Manindra Agrawal, IIT Kanpur’s director, described the program as being meant especially for young ethical hackers and said admission would happen through a hackathon. somebody replied asking what material they should prepare from. his response was perfect: “if you need to prepare for it, the program is not for you.”

that is probably one of my favourite things an indian university director has said about education because that is the entire fucking point. the thing you want to select for is the kid who was already hacking when nobody offered marks for it. the one who learned because not understanding something annoyed them. not somebody who discovered cybersecurity three weeks ago because a coaching channel uploaded COMPLETE ETHICAL HACKING ROADMAP 2026 🔥.

B.Cyber is not perfect. JEE Main is still part of screening⁵. apparently even when designing a degree specifically to find hackers, india still needs a little PCM foreplay before the actual hacking begins. but it is directionally much better than pretending AIR is a security aptitude test, and i would love to see indian technical education move much further in this direction.

because, in my experience, most IIT CTF teams are honestly pretty shit. there are exceptions, obviously, and CTFs measure only a narrow slice of security ability, but the idea that putting very high-JEE-rank people in one place automatically produces elite hackers is just funny. as of september 2026, none of the top 15 teams in CTFtime’s india ranking are IIT teams; InfoSecIITR appears at #16, while z0d1ak is #1 in india and #10 worldwide, BITSkrieg is #3 and bi0s is #5⁶. fifteen teams sit above the most prestigious engineering brand in the country, and not one of them got there on rank. remove the institutional signalling, put a challenge on the screen and the prestige hierarchy stops reproducing itself.

the machine wants the flag. it does not want your AIR.

the same problem becomes much larger once you move from cybersecurity to AI and frontier engineering. india does not lack intelligent people. it lacks enough environments where intelligent people are given absurd amounts of compute, ownership, capital and permission to move fast. good AI research now increasingly requires GPUs, systems engineering, data, infra and the ability to run expensive experiments repeatedly. publishing another paper is not the same thing as putting a small group of cracked researchers next to a giant cluster and letting them try stupid ideas until one stops being stupid.

this is where the gap with the united states becomes painful. american labs and companies operate with a density of compute, capital, researchers and technical ambition that is still difficult to reproduce here. Stanford’s 2026 AI Index says industry produced more than 90% of notable AI models in 2025, with the US producing 59 notable models that year and massively leading private AI investment⁷. india itself clearly recognizes the compute problem: the IndiaAI Mission created a national compute pillar built around expanding access to large-scale GPU infrastructure⁸.

we need much more of that, and not only giant centrally managed facilities. a sufficiently good nineteen-year-old should be able to have a stupid idea at 2 am, ssh into something enormous and find out whether it is stupid by breakfast. researchers should not spend more energy negotiating access to compute than using it.

this is probably the part of san francisco culture i am most jealous of. SF and the broader american tech ecosystem are extremely comfortable giving weird, young, obsessive people disproportionate resources when they show disproportionate ability. there is far less psychological need for somebody to wait until their age, degree or designation catches up with what they can already do.

that is the part of the a16z / palantir worldview i am completely bought into. find exceptional people. increase talent density. put them frighteningly close to real problems. give them GPUs, source code, capital and responsibility. collapse the distance between competence and ownership. if the twenty-year-old is the best person to own something important, give it to the twenty-year-old.

Palantir explicitly describes advancing successful hires into major responsibility based on merit even when they have limited previous industry experience⁹. a16z increasingly talks about conventional resumes becoming weaker signals in an AI world and the need to identify exceptional people through better evidence¹⁰. this is such a radically healthier instinct than “you aren’t in third year yet.”

signal over pedigree. output over tenure. agency over procedure.

that is what india should copy from silicon valley, not startup vocabulary and linkedin aesthetics. talent density. technical ambition. compute abundance. aggressive delegation. taking young people seriously. being willing to bet heavily on someone because the evidence says they are exceptional, not because the correct institution has stamped them enough times.

the best technical people are often weird. obsessive. disagreeable. autistic about some tiny technical thing nobody else notices. they spend fourteen hours inside vllm because one scheduling decision bothers them. they rewrite a renderer five times. they read some forgotten mailing-list thread from 2007 because it might explain an optimization. they see an undocumented endpoint and lose the rest of their evening.

you don’t manufacture these people through procedural compliance. you find them, give them hard problems and avoid suffocating them.

instead india often takes exactly the person who should flourish — curious, impatient, deeply technical, somewhat incompatible with bureaucracy — and puts them through years of examination fetish, seniority fetish, credential fetish and permission seeking before they’re trusted with anything interesting. then some of the strongest people leave for places with stronger peers, better labs, more compute, more capital and a culture that lets them move much faster.

then we call it brain drain.

maybe stop draining the fucking brain while it is still here.

AI is going to make the credential fetish even more embarrassing. intelligence is getting cheap. codex can already produce the currency converter, crud backend, portfolio, documentation and enough github slop to impersonate a mediocre engineering student indefinitely. mechanically producing software is getting cheaper. mechanically producing evidence that you “know programming” is getting cheaper too.

what becomes expensive is judgement, taste, agency and obsession. can you choose a problem nobody assigned you? can you read source when the abstraction breaks? can you notice one strange response among ten thousand normal ones? can you tell when the model is confidently bullshitting you? can you benchmark your beautiful idea, discover it sucks and delete three days of work without having an ego crisis?

this is why mediocre egoistic people are so exhausting. ignorance is fixable. inexperience can disappear frighteningly quickly. ego is different. you cannot teach somebody who experiences correction as humiliation. you cannot build frontier systems with people whose first instinct is defending their status. you cannot have a meritocracy where age, degree, institute and designation remain sacred cows.

age is a proxy. CGPA is a proxy. JEE is a proxy. GATE is a proxy. codeforces is a proxy. your college is a proxy. increasingly, even your resume is a proxy.

the thing itself is competence.

give me five people who attack unfamiliar systems until they understand them over fifty people waiting for jira to tell them what to think next. give me the freshman reading vllm internals. give me the teenager reversing binaries at 3 am. give me the person rewriting the renderer for the fifth time because version four still feels wrong.

give them GPUs, root, source code, impossible problems and an unreasonable amount of autonomy.

then get out of their fucking way.

the future belongs to high-agency autists.

Footnotes

1. JEE Advanced 2026 requires indian candidates to qualify through JEE Main; IIT admission through JoSAA requires a JEE Advanced rank. JEE Advanced eligibility / admission criteria. ↩

2. rishabhxchoudhary, why is there so much cheating in contests in india? especially on codechef?, codeforces. the author describes live solution sharing and connects some demand for ratings to resume shortlisting. ↩

3. codeforces community posts have documented accusations involving leaked solutions and students associated with IITs. these are community allegations rather than institutional findings. ↩

4. CERT-In, Scientist-B recruitment 2026. candidates were shortlisted using valid GATE scores; final merit assigned 60% weight to GATE and 40% to interview/interaction. official advertisement. ↩

5. IIT Kanpur B.Cyber admissions consider JEE Main performance plus evidence of prior cybersecurity work for shortlisting, followed by a six-hour hackathon covering cryptography, web/API/network security, reverse engineering and osint. IIT Kanpur. ↩ ↩²

6. CTFtime India standings, accessed september 2026. ↩

7. Stanford HAI, 2026 AI Index, research and development and economy. ↩

8. the IndiaAI Mission’s compute pillar was built around making supercomputing capacity of over 10,000 GPUs available to startups, researchers and academia. IndiaAI, cabinet approval. ↩

9. Palantir describes advancing successful hires into substantial responsibility based on merit despite, in some cases, limited prior industry experience. Palantir 10-K. ↩

10. a16z’s Talent Engineer Fellowship discusses traditional resumes becoming noisier signals in the AI era and finding exceptional candidates through stronger signals. a16z. ↩