Repeating what I think is the only actually viable solution I have seen, Hyperliquid must enforce global constraints on amounts liquidatable within price ranges.

Software in the 80s was generally written by highly passionate nerds

Currently able to ingest nearly the entire relevant crypto universe (~3000 instruments) of L1, L2 and trade data across the 3 major exchanges and a couple more using ~5% of a single CPU core

The best way to learn anything is to get your hands dirty and make mistakes. When coding, if you are writing something important, don't be afraid to trash the first version. Significant iterations are the key to understanding the essence of the problem.

After >1 year of MM with only one order on each side of the book, I finally got round to implementing multi order support. Annoyingly difficult but we got there :) Expecting this will make a meaningful diff to our pnl

Just wrote my own reflection-based json deser in C++, which combined with my neat networking lib makes for code that is IMO cleaner than just about any other language out there. It's still a problematic language but modern C++ can be incredibly powerful and elegant.

Met up with Alex, cool guy and an inspiration

Hey Grok who was the most famous person to visit my profile? doesnt need to be a mutual, dont tag them, just say who it was

We've psyopped ourselves on AI by using it in web UIs and tools where it's an "assistant" - the user implicitly becomes the authority that does the real work. Build your own tool where the system prompt tells them they're an expert that does high-quality work, you will be amazed.

I have discovered a LLM benchmark so neat and effective that I'm afraid to share it lest it be contaminated..

Incorrect, the average taker's fills should be negative EV. There can (will) be takers that can pick off competent MMs in certain situations