Hi, I’m Mithil.
I graduated from IIT Bombay with a B.Tech in Engineering Physics.

I do research in AI.
I want to solve the sample efficiency problem
This should directly lead to superintelligence

I recently showed that small transformers can perform extremely well on ARC-AGI, a famous metalearning benchmark that is great for sample efficiency.

These results got attention from famous researchers and went viral on X, since many people considered it impossible previously. My model beats many LLMs, despite needing only 67 cents to train it (from scratch!) and is likely the pareto frontier.

This was my attempt to test the limits of today’s deep learning methods. Now I’m working on newer ideas. I expect to beat every LLM soon

Here’s my blog

Here’s some cool shit I’ve done:

• Team India selection camp for the Astronomy Olympiad without studying
• Cracked JEE in an extremely unconventional manner
• Biohacked my way out of a misdiagnosed health condition
• Saved myself in a medical emergency (click dropdown for story)

  • My biceps got ridiculously swollen after a bad gym session. The pain was excrutiating, 10x worse than any DOMS i felt before and I couldn’t move them at all. I had learnt enough medicine during the biohacking to know it was rhabdomyolysis, which is life threatening.
  • I remembered that confirmation requires a blood test that measures CK values. Went to a private lab, ordered one - my values were 200x the upper limit and rising (diagnostic confirmation + requires immediate hospitalisation). The pain was high enough now that I was also worried about compartment syndrome
  • Went to the nearest ER with my report and my extremely painful, swollen arms. (This was a mistake, the avg indian hospital is filled with untrained staff. I should have gone to the best hospital in the city)
  • The doctors were dumb and didn’t know basic stuff. She had never heard of CPK or rhabdo or compartment syndrome! She called a senior ortho consult. He didn’t know either! They refused treatment, saying “go home, nothing to worry about. We’ll check tomorrow”. Called the next closest ER - same story. They hadn’t heard of it too!
  • I freaked out. I knew I could potentially lose my kidneys if I didn’t get treatment right away. Ideally going to a nephro would solve it since they would clock it immediately but it was late at night and I didn’t know any Nephrologists.
  • I immediately began a lot of ORS as a proxy for saline to protect kidneys till I figure something out. But I knew I needed hospitalisation immediately
  • An idea struck. I realised CK values can go high during any muscle cell death scenario, includin a heart attacks. So a cardiac must know about rhabdo!
  • I went to the cardiac ER of the same hospital, I needed some excuse to get in so I told the nurses I had chest pain (i had mild soreness)
  • I finally got in front of a cardiac but they didn’t know about rhabdo either! (yeah fml, next time only go to the best hospital). However they took me in coz the report made them worried about a heart attack
  • I fought for a few hours, saying “screw the chest pain, call a nephro”. They finally relented and rang him at 3am. He told them over call that I was right and that its obviously rhabdo, and gave the correct treatment plan
  • The nephro (who came the next day) said he didn’t like my actions but it might have saved my kidneys, or even my life. He was stunned that i recognised rhabdo this early AND got the CK test done myself. He wasn’t sure if an actual doctor would have clocked it this early

Stupid/fun shit:

• Got stuck solo in a random French city at midnight without a room
  • Greatest night of my life
• Slid down a slope in the swiss alps taking a shortcut on a trek
• Ended up in Italy accidentally due to altitude sickness (unrelated to above)
• Accidentally deleted an employee database of Infosys as a kid

I casually play soccer, badminton, snooker, squash, and the drums.
HMU if you’re playing or jamming in blr

xeets @evilmathkid
LinkedIn for the cv
mvakde [11 + 1] at gmail