When the stock was down in the dumps, this day seemed unimaginable. But the team kept grinding, regardless of market conditions, and now we’re finally seeing the rewards. Huge congrats and a big thank you to everyone at @RobinhoodApp
Interesting thing @mayankag and I observed today. If O1 ever starts answering "A succinct way to see..." you are screwed, the model always hallucinates after this.
Earlier today was INMO (Indian National Math Olympiad) 2025. As a case study, I wanted to see how far one can go in solving NEW math olympiad problems by just using LLMs.
Here is a solution set that I was able to produce by just guiding the state of the art models (like o1, deepseek etc). I have not seen an official solution and none exists to the best of my knowledge.
While I am not convinced that all 6 answers are correct, I am very confident that most are correct or almost correct. I learned a lot from this exercise and will perhaps write a separate post about it.
Its crazy to imagine what the future would be like. Who could have thought that a computer would be able to solve problems like these…
Question Paper: https://t.co/VvSAnms2d2
AI (+ Human Nudged) Answers: https://t.co/oJdFjBVEaN
We were benchmarking O1-pro on some math Olympiad problems, and it got the attached problem wrong. But when we probed further, it produced elegant counterexamples proving it was right—and the solution I was using were wrong! Kind of blown away.
New email from Elon Musk to engineers: “please be prepared to do brief code reviews as I’m walking around the office.”
That’s it — that’s the whole email.
The future of investing is 24/7. @RobinhoodApp is working towards offering 24/7 equities trading, and today, we’re taking a step towards that goal. (1/9)