
I think the explosion of “neolabs” in quantitative trading has been brought about by a misconception either in the competitive landscape OR where money is made in quantitative trading. The idea that a startup claims to be the first to pursue “the bitter lesson” in quantitative trading is absolutely ludicrous. Top tier quant funds have long been hiring AI researchers and training end-to-end (industry parlance: tick-to-trade) models for years. From memory: HRT, UbiQuant, XTX, HighFlyer (hello, deepseek?!), and many, many, many more. In every podshop, there is at least 1 team with a focus on creating large scale end-to-end models that “train on markets directly”. If you CLAIM to be the first, you either don’t understand the landscape or are straight up lying, or don’t even have the slightest clue on the SOTA problems being worked on by your past employers. — I can’t speak to their skills, but there is some insinuation here that training the largest model and throwing a lot of data at the model will “solve” quantitative trading. Markets are for better or worse, reflexive. That makes it hard for anyone to achieve high predictive power for a long time. Research is almost always a constant race to maintain your position - unlike more general machine learning, there is no evidence that SoTA models in markets have been able to continuously improve their predictive power beyond a certain threshold. The implication here is that money is often made not purely from the models, but also the non-trivial technological and operational scaffold of being able to monetise low signal-to-noise models. Spoken plainly, there is a lot of work to be done across the entire investment process in order for a single dollar to be generated. When one has been a fledgling researcher, he is tasked only to focus on and think about the model. And upon any initial success, he is almost always convinced that his portion is the lion’s share of the work, oblivious to the thousand moving parts that made his model work.
Today, @bayesiankalki and I are launching Mnemonic Labs (YC F26). We're a neolab training a foundation model for quant trading that learns directly from market activity. In <6 weeks we’ve built our training pipeline and model architecture. Our current training run uses over 25TB of tokens. We met 28 years ago shortly after I was born (we’re brothers!). We’re Olympiad finalists (5x USACO combined) and MIT graduates. Kalki made $20M in annual PnL at 22 at Tower Research and I was the youngest quant at Citadel at 19. Over the past ten years, we’ve been part of quant trading’s evolution from linear models to trees and now neural networks. Learn more about us at ycombinator.com/launches/XRK… or reach out at founders@mnemoniclabs.ai!










