April 28, 2026

TL;DR. Rishabh Agarwal's ICLR 2026 talk: scaling RL on 1T+ parameter frontier LLMs using experimental data from Periodic's physical lab.

Takeaways

• Talk recording and slides linked in the thread for anyone wanting the full technical context
• Periodic is running RL experiments on frontier-scale models (1T+ params) grounded in their own physical lab data
• Team explicitly signals they're scaling further — worth tracking if you follow frontier RL work

TL;DR. Quant firms like Jane Street and Citadel pay $336K–$1.4M because they hire math and CS brains, not finance majors — and the path in is a structured probability/stats stack, not market intuition.

Takeaways

• Top firms explicitly don't require finance backgrounds; 2/3 of Jane Street's recent intern class studied CS or math.
• Four roles to know: Quant Researcher (highest paid), Quant Trader, Quant Developer, Risk Quant — with ML/AI roles growing fastest.
• Learn in order: probability (Bayes, EV) → statistics (hypothesis testing, signal vs. noise). Skipping foundations kills you in interviews and live trading.
• Prediction markets are the new frontier — same systematic frameworks as equities, applied to elections, sports, and geopolitics.