"AI people don't like C++," Jane Street has its traders use this language instead
We first highlighted Mojo as a language to watch in 2023; it's designed to work on any hardware and be superfast while being written in 'pythonic code'. As it continues to catch on, top firms are taking notice. Electronic trading giant Jane Street is one of them.
The language was created by AI firm Modular, led by ex-Google senior director Chris Lattner, who was recently invited to appear on Jane Street's Signals and Threads podcast. Lattner said "AI people don't like C++" because it "has pointers, and it's ugly, and it's a 40-year-old language." Python is by-and large the language of choice in machine learning today due to libraries like PyTorch and TensorFlow, and imitating its syntax makes it much more accessible to those people when trying to maximize performance.
A key strength of Mojo, Lattner said, is its metaprogramming capabilities (i.e. its ability to modify code while a program is running to optimize it). In the context of GPUs, he said that the tensor cores of an NVIDIA chip are very different to one from AMD, and a metaprogramming allows you to pick the best code for whichever chip you use without having to manually change the code each time. The same thing is possible through C++, but Lattner said errors in the code will create "some crazy template stack trace that is maddening and impossible to understand." Mojo provides a "very simple error message" allowing for easier debugging. Jane Street CTO Ron Minsky, who hosts the podcast, said the firm is also working on metaprogramming features for OCAML.
Jane Street has its traders use PyTorch when working in machine learning because of how easy it is to iterate through different strategies until an effective one is found and optimized. PyTorch is often put through an AI compiler that Lattner said will "generate optimal code for some chip" and "magically get out high performance," but the issue with this is that you won't know whether you've truly got the most optimal code. Jane Street runs many medium/low frequency trading strategies, where that last ounce of performance doesn't matter as much, but Mojo can provide a big impact for its high-frequency teams. The firm has more than 5,000 GPUs; it might as well make the most of them.
Minsky said that traders have started using Mojo now as well and said it has "a lot of exciting things going for it." In addition to the performance boon, Minsky said Mojo has a level of "explicitness [that] is actually really important for performance engineering."
Relying on magic has worked for Jane Street so far, however. It made a record-breaking $10.1bn in net trading revenue in just one quarter earlier this year. It made $10.6bn during the entirety of 2024, for reference.
Have a confidential story, tip, or comment you’d like to share? Contact: Telegram: @AlexMcMurray, Signal: @AlexMcMurrayEFC.88 Click here to fill in our anonymous form, or email editortips@efinancialcareers.com.
Bear with us if you leave a comment at the bottom of this article: comments are moderated intermittently by human beings. Sometimes these humans might be asleep, or away from their desks, so it may take a while for your comment to appear. You must take sole responsibility for comments you post on this site. We will take reasonable steps to weed out anything that we consider to be offensive or inappropriate.
