i am on my knees begging you to read this, it can literally get you a $300k quant job
it turns Grok Bot into a strict tutor that drags you through FREE Harvard, MIT and Stanford courses and never lets you quit
steal it before everyone finds out ↓ https://x.com/i/article/2117685410373517313
Grok Bot: How to Become a $300K Quant With an AI Tutor
44,055 people bookmarked the most famous quant roadmap on this app. I was one of them, and I lasted 4 days.
A quant is the person at a trading firm who uses math and code to decide what to buy and sell. The roadmap opens with the pay: in 2025, new grads at top firms made $300K to $500K in their first year.
You can learn all the math on that roadmap for free, from books and lectures by Harvard, MIT and Stanford. I had everything except one thing: someone who checks whether I did the work.
So I built that someone in Grok Bot. My tutor sends me 3 problems every morning, never gives me the answer first, remembers every mistake I make and won't let me skip a level.
The fix was so simple that I'm a little embarrassed I needed a failed attempt to find it.
34 days later, level 1 is done, 212 problems are solved, and I haven't skipped a single morning.
Inside:
- why 97 out of 100 people quit free online courses
- the 1984 study that shows why a tutor beats a normal class, and the one rule that turns AI into a tutor instead of an answer machine
- how to build your own tutor in Grok Bot in 20 minutes, with every prompt
- the 5 levels of quant math in plain words, with free books and lectures from Harvard, MIT and Stanford
- my numbers after 34 days, and everything that broke
- my tutor as a template you can add to Grok Bot in one click
Bookmark this and build it tonight, so your tutor can send you the first 3 problems tomorrow morning.
Part 1. Why almost everyone quits
01. The content was never the problem
MIT and Harvard have put full courses online for free for years. Researchers looked at 12.67 million sign-ups to those courses over six years. In the last year of the data, 2017-18, only 3.13% of the people who took part finished. That is about 3 people out of 100.
The quant roadmap hit the same wall. The biggest quote post under it was called "YOU WILL NOT BECOME A QUANT" and got 605,931 views. Its point: all the content is free, and you still won't make it, because you won't grind through 200 problems in a row.
I didn't make it either. These are the 4 reasons I quit, and they will hit you too.
02. Nobody checks
Skip a day and nothing happens. Skip a week and nothing happens either. A PDF never asks where you went.
03. "Got it" lies to you
You read a definition, it makes sense, you say "got it" and move on. In a 2019 study, students who had to solve problems in class learned more than students who sat through a polished lecture, and they felt like they had learned less. Your feeling of understanding is a bad judge of what you know.
04. The chatbot gives you the answer
In a study of about 1,000 high school students, the group with ChatGPT did 48% better on practice problems. Then came the exam without AI, and they scored 17% worse than students who never had it.
05. The chat forgets you
Tomorrow you open a new chat and start from zero. It doesn't know you got conditional probability wrong 3 times this week. If you don't open it, it never comes looking for you.
So the missing piece is someone who checks on you every day and makes you do the thinking yourself.
Part 2. Why a tutor works
06. The student who beats 98% of the class
Alexander the Great had Aristotle as his private tutor. For most of history, one-on-one teaching was something only the rich could pay for.
In 1984 the education researcher Benjamin Bloom measured what it is worth. Students taught one-on-one did better than 98% of students taught in a normal class. Later studies found a smaller effect.
Bloom called it the 2 sigma problem: tutoring works, and almost nobody can afford it. A math tutor costs $46 an hour on average. Two hours a day for a month comes to $2,760.
07. AI tutors already pass the test
- Harvard, 2025: 194 students. The ones who studied with an AI tutor learned more than twice as much as students in an active class, in 49 minutes instead of 75.
- World Bank, Nigeria, 2025: six weeks with an AI tutor (English lessons, with teachers in the room) gave gains the researchers compare to 1.5 to 2 years of normal school.
08. The one rule that decides it
Back to the study with 1,000 students. It had a second AI group, and their ChatGPT gave hints instead of answers. They did 127% better on practice problems, and on the exam they did no worse than students without AI.
Same study, same AI. The difference was a single rule: no answers.
That rule also trains the exact skill quant interviews look for. The roadmap points out that Jane Street gives problems that are too hard to solve alone, on purpose, to see how you use hints.
Part 3. Build your tutor in 20 minutes
In Grok Bot you can create bots: AI teammates with a name, a job, their own chat and a memory that grows over time. Bots work on a cloud computer, so yours can read files and write and run Python. Routines run in the cloud on a schedule, so your bot messages you even when your laptop is closed.
That covers all 4 reasons from Part 1:
- nobody checks → it messages you every morning
- "got it" lies → it tests you from memory every evening
- the chatbot gives answers → its first rule puts hints before answers
- the chat forgets → it keeps a log of every mistake
09. Step 1. Create the bot and give it rules
Create a new bot, name it Quant Tutor and send this as the first message. Keep the last line: it tells the bot to save the rules to its memory.
Then check that the rules stuck. Mine answered "Got it. Rules locked in." When I asked it to list its memory, the 10 rules weren't there. A "got it" from a bot needs a test too.
If the 10 rules aren't on the list, tell it to save them. Mine noticed the gap by itself and saved them on the spot.
10. Step 2. Give it the free courses
One source per level:
1. Probability: Introduction to Probability by Joe Blitzstein and Jessica Hwang. The authors put the second edition online for free at probabilitybook.net, and the Harvard Stat 110 lectures are on YouTube.
1. Statistics: MIT 18.650 Statistics for Applications, with lecture videos and slides free on MIT OpenCourseWare.
1. Linear algebra: MIT 18.06 with Gilbert Strang and his video lectures.
1. Optimization: Convex Optimization by Stephen Boyd and Lieven Vandenberghe, a free PDF from Stanford.
1. Stochastic calculus: MIT 18.S096 Topics in Mathematics with Applications in Finance. Lecture 18 is Itô calculus, and lecture 19 is the Black-Scholes formula.
The roadmap uses paid books for statistics and stochastic calculus. I swapped them for free MIT courses on the same topics.
Keep the week budgets in. Without them my tutor planned the full courses: 59 weeks, about 14 months. With them it came back with 25 weeks.
11. Step 3. Take a placement test
Mine found the holes in my school algebra on day 1 instead of week 2.
12. Step 4. The morning routine
Check the time zone in your Grok Bot settings first, then send:
Right after the rules, mine had already set up a 9:00 morning check-in on its own. This prompt moved it to 8:00 and added the 1:00 PM reminder as a second routine.
13. Step 5. The evening recall
Why from memory: in a 2006 experiment, people who tested themselves remembered 61% of a text a week later, more than the people who simply reread it.
14. Step 6. The Sunday exam
15. Optional. A study group
Grok Bot can put several bots in one group chat with shared context. I added two: an Examiner that only writes the Sunday exams and a Coach that only watches my streak.
That is the whole build: one bot and four routines, about 20 minutes of setup. My tutor kept growing after that, and the template in Part 7 is its current version.
Part 4. The 5 levels in plain words
The order comes from the roadmap, and you can't skip a level, because each one stands on the one before it. The weeks are its own estimates at 2 hours a day: 21 to 28 weeks in total, so about six months of math.
You don't need a finance degree. According to the roadmap, over two-thirds of a recent Jane Street intern class studied computer science, over a third studied math, and finance knowledge generally isn't required.
Each level ends with a boss: a question in the style of a real quant interview. The answers are at the end of this part.
16. Level 1. Probability: how likely is it? (3 to 4 weeks)
You learn to count outcomes and to update your guess when new facts come in (that is the Bayes rule). The roadmap says to solve every problem in chapters 1 to 6 of Blitzstein and Hwang. After you guess, your tutor flips a coin 10,000 times on its computer and shows you who was right.
Boss: You roll two dice. A friend peeks and says "at least one of them is a 6". What is the chance that both are 6s?
17. Level 2. Statistics: was it luck, or does it work? (4 to 5 weeks)
You learn to tell a real pattern from noise, which is where a lot of trading ideas die. Your tutor pulls real stock prices and makes you check whether big crashes happen more often than the textbook formula predicts.
Boss: You test 1,000 trading ideas that are pure noise. The standard test lets 5% of junk through. How many of your ideas will look like they work?
18. Level 3. Linear algebra: math for many things at once (4 to 6 weeks)
A portfolio is hundreds of prices moving together. Linear algebra lets you handle all of them at once and find the few hidden forces that move most of the market. The roadmap calls the Strang lectures non-negotiable. Your tutor turns each lecture into 3 problems for the next morning.
Boss: You follow 500 stocks. How many separate numbers do you need to describe how much each one swings and how every pair moves together?
19. Level 4. Optimization: finding the best choice (4 to 5 weeks)
This is how you decide how much of each stock to hold, so you get the most return for the risk you can live with. Your tutor writes the code for a small portfolio, and you predict the result before it runs.
Boss: Two stocks each swing about 20% a year, and they move independently of each other. How do you split your money so the total swings the least, and how much does it swing then?
20. Level 5. Stochastic calculus: math for prices that move at random (6 to 8 weeks, the hardest)
School calculus works on smooth curves, and prices jump around. This level teaches you the rules for doing math on random moves, and it ends with you deriving the Black-Scholes formula yourself. That is the formula that prices options: contracts whose value depends on where a price goes.
Boss: A stock rises 10% on Monday and falls 10% on Tuesday. Are you up, down or even?
21. Answers
1. 1/11. Of the 36 ways two dice can land, 11 include at least one 6, and only one of those is a double 6.
1. About 50, since 5% of 1,000 pass by luck alone. This is why a strategy that looks great on old data can fall apart with real money.
1. 125,250: 500 numbers for how much each stock swings, plus 124,750 for every pair.
1. Half and half. The total then swings about 14%, which is 20% divided by the square root of 2. Mixing two risky things gave you less risk than either one alone.
1. Down 1%: $100 becomes $110, then $99. The swings ate a dollar even though the average move was zero. That lost 1% is the extra term in the Itô lemma, the main rule of level 5. Proving it is part of your homework there.
Part 5. One day with my tutor
Day 19 went like this.
The morning message came with my streak at the top and 3 problems. Problem 2 came straight from my mistake log:
1% of people have a rare disease. A test catches 90% of sick people and wrongly flags 9% of healthy people. You test positive. What is the chance you are sick?
I typed "90%".
Tutor: Wrong. You skipped the 99% who are healthy.
I stared at that for a while and asked for a hint.
Tutor: Hint 1: picture 1,000 people. How many are sick, and how many of them test positive?
10 are sick, and 9 of them test positive. Then the healthy ones: 990 people, and 89 of them get flagged by mistake. So 9 real cases out of 98 positives, about 9%.
Tutor: Right. Now tell me why it is so low when the test catches 90%.
It took me 10 minutes to say it in plain words: healthy people are so many that even a small error rate produces more false alarms than real cases. The tutor added the mistake to my log and scheduled it to come back on days 20, 22, 26 and 40.
That evening I got 5 questions and no notes. I scored 3 out of 5, and both misses went into the log.
Two days later came the Sunday exam, and I passed it with 85%. The weekly report listed my weakest topics: conditional probability and counting.
Part 6. My numbers after 34 days, and what broke
22. The numbers
- Streak: 34 days, 0 skipped mornings
- Problems solved: 212
- Level 1: done on day 26 (the roadmap estimate is 3 to 4 weeks)
- Evening recall: 2.1 out of 5 in week 1, 4.4 out of 5 in week 5
- Sunday exams: 3 passed, 1 failed (week 2, so I repeated the week)
23. Where it went wrong
1. In week 1 the tutor leaked answers. When I pasted a wrong attempt and asked it to "check this", it checked by writing out the full solution. That is where rule 2 came from: say "wrong", point to the step, wait. It hasn't leaked since.
1. My confidence lied to me. In week 2 I told the tutor I was 9 out of 10 sure I understood the Bayes rule. That evening I scored 2 out of 5. Researchers at Wharton found the same pattern with AI in 2026. When the AI was right, people scored 25 points higher. When it was wrong, they scored 15 points lower, and their confidence went up anyway. They call it cognitive surrender, and it is the reason my tutor never answers first.
1. It is still 2 hours a day. The bot doesn't do the math for me. It makes sure I show up, and it makes skipping feel worse than doing the work.
1. I haven't reached level 5. It is 6 to 8 weeks of the hardest math on the list. I'll report back from there.
Part 7. Take my tutor
https://x.ai/bot/eqsjxgqMhiNo0U03FDmda
Open the link and press "Add to Grok Bot". You get the tutor from Part 3, upgraded. I kept adding to mine, and this is what it has now:
- 14 rules instead of 10. The new ones make it ask how sure you are before it grades you, mix old topics into every set, keep you at 70 to 85% correct and make you derive the key formula of each level from memory before you move up
- 11 skills, step-by-step playbooks it follows for each job, including the placement test, the morning set, hints, the evening recall, the Sunday exam with a report, a Friday simulation lab, a monthly mock interview and a market-making game where you quote prices and it trades against you
- 6 routines: problems at 8:00 AM, a nudge at 1:00 PM if you haven't answered, recall at 9:00 PM, the exam on Sunday, the lab on Friday and a mock interview on the first Saturday of every month
- the 25-week plan with the 5 free sources, and what comes after it: 5 famous research papers on trading, each one rebuilt in code from a blank file
In the first chat it sets itself up. It saves its rules and plan to memory, asks about your goal, your background, your hours and your time zone, and starts a 10-question placement test. Before it grades an answer, it asks how sure you are. At the end you see where you start and how often you were right when you said 90%.
Then:
1. Check the time zone in your settings, because the routines run on it.
1. Ask it to list its memory. If the 14 rules aren't there, tell it to save its Tutor rulebook skill to memory.
1. Tomorrow morning, solve the first 3 problems.
Would you rather build it yourself? Every prompt is in Part 3.
The tutor from Part 3 isn't tied to quant math. Swap the 5 sources and it teaches Spanish, Python, chess openings or the bar exam with the same rules.
Part 8. What comes next
Math gets you to the interview. The next article is the interview tutor: the Green Book (200+ real interview problems), Zetamac mental math drills (the roadmap target is 50+), the Jane Street monthly puzzles and Figgie, the trading card game from Jane Street.
It comes out the week I pass the level 2 exam.
Follow me so the next article lands in your feed, and bookmark this so the prompts are here when you start. What would you make your tutor teach you? Tell me in the replies.
44,055 people bookmarked the roadmap. Be one of the few who finish it.