Evidence of Exceptional Ability
1) My work on ARC-AGI
Results I open sourced
I built a tiny transformer that scores 44% on ARC-AGI, trained from scratch in just 67 cents.
- This is the best non-LLM method in the world
- Also the cheapest, and trains the fastest by far (new pareto frontier)
- Beats every single non-thinking LLM and many thinking ones
- Least # of params at this performance
- Went viral on X, got the attention of top AI researchers (many thought the result to be impossible initially)
All the posts about my ARC work are on X: https://x.com/evilmathkid/highlights
More details in Blogs: 44% on ARC-AGI 1, A New Pareto Frontier on ARC-AGI.
Code (open source): https://github.com/mvakde/mdlARC
Results I haven’t released yet
Since then, I have doubled the score again and saturated ARC-1 (cost remains incredibly low).
Now working on v2 of the benchmark, currently at 21% (beats GPT-5 Pro) 27% now and beats Opus 4.5 Thinking (16k) I expect to beat the WR (~40%) in 2 weeks.
Winner of v2 gets $425K (spoiler: its gonna be me)
Why ARC?
IMO sample efficiency is the most important problem in AI today.
ARC is a great benchmark for it: v few examples per task, diff rules for each task, v few tasks.
I was using this benchmark to test the limits of transformers. Now I am working on creating new methods better than AR transformers.
Bragging points:
- I learnt ML almost from scratch doing this, and in a v short period of time. A year ago I didn’t know how transformers worked. Today I am pushing the frontier on a benchmark that has been open since 2019 (Despite fame/heavy cash prize, researchers made no major progress till O1/O3 in 2024)
- The only models with comparable performance are frontier reasoning LLMs, which need millions in pretraining compute. My model’s lifetime cost is just a few dollars
- I did this alone with no advisor
- The compute costs of all my experiments over the past few months is only about a 1000 dollars
2) Topped extremely competitive National Exams without studying much
Top 25 all India, Indian National Astonomy Olympiad
- This got me invited to the Team India selection & training camp (International Olympiad in Astronomy & Astrophysics)
- I did this without studying at all. I was already among the best in the country at physics and had some general knowledge on Astronomy
Ranked 1500 out of 1.2 million starting candidates in JEE Advanced
- Almost everyone else who qualified spent 2-4 years studying 10+ hours a day, and practice 10s of thousands of problems
- I barely practiced any questions (< 10 a chapter), skipped 2/3rds of chemistry (and all the memorisation heavy parts of math/phy), studied a only a few hours a day, for only about a year. Difference was I spent all my time on understanding the theory at a ridiculous college lvl depth. More details
Career deets
Education
B.Tech @ IIT Bombay, 2019-2023 Majored in Engineering Physics
Academic Achievements
- NIUS Fellowship by Tata Institute of Fundamental Research, 2021
- Ranked 324 all India in KVPY. Got the fellowship twice in 2018 & 2019.
- Top 300 all India, National Standard Exam in Astronomy
- Runner up, IBM teen hackathon (I was in 9th grade, everyone else in 12th grade)
Work Ex
1. Trying to solve sample efficiency Nov 25 onwards
Started this after spending a few months learning ML
IMO solving sample efficiency directly leads to superintelligence
Tested new techniques on ARC and got some insane results
2. Founder, Healthcare AI startup Oct 24 - Apr 25
Explored building in Healthcare+AI. Iterated over many MVPs, most failed, one got some traction.
Dropped this when I realised I wanted to build AGI
3. Founding team, GetCrux (YC W24), Dec 23 - Sep 24
Led a core product - Automating insights for performance marketers
Was a backend dev and product manager on other projects.
Led many project verticals till we got our first non-tech hire - sales, marketing, research for pivots, prototyping, recruiting, etc. (I was employee #3 and the only generalist)
4. Research Intern, University of Paris Saclay May 22 - Jul 22, in-person
Found a correlation b/w DNA content & light absorption while analysing UV images of viruses.
Did a bunch of physics experiments on RNA/protein mixtures
5. Intern, New ventures arm @ Tata Mar 23 - Jun 23
Evaluated business models / industry research for Deeptech investments.
Regularly presented to senior leadership of Tata Industries and other companies