The Exciting Mojo Language

•

main.🔥 Having the new Mojo language files end with a .🔥 emoji is just the cherry on top of the incredible promises @Modular_AI has just dropped on the world. Watched the keynote, and a few breakdowns, it's incredibly exciting vimeo.com/822336980/20533747fa

Mojo is being developed by @Modular_AI under the supervision of @clattner_llvm who was behind @SwiftLang and LLVM and the MLIR compiler. Chris shines during this keynote!

Claiming performances boosts of up to 35,000x! over python, Mojo is a superset of Python, and is aiming to be fully compatible with the ecosystem and packages! This is immense! Just imagine how much this adds to the two exponential AI curves (GPUs x arXiv submissions)

You can think of Mojo as Typescript, that adds not only types, but incredible speed boosts and obvious benefits, to make developers want and making it easy for them to switch. And the plan is to Open Source Mojo in due time.

The full keynote (and may I add, it was incredible in it's production value, not shaming Apple!) goes deeper into the types of improvements Mojo adds, borrowing from Rust, Swift and other languages for the performance boosts and scalability.

Mojo is introduced along side the @Modular_AI AI Inference engine, that also claims extraordinary things. Being written in Mojo, they claim to be able to run (via MLIR) on all infrastructures, CPU, GPU, TPU, all the PUs (their joke, not mine)

They claim drop-in compatibility, with no plugins or tricks, to make this run on any architecture. I... I want to see how this shakes out because this has huge implications to costs, speed of every new model demo, and tons more.

They are talking about how the current AI infrastructure is built on tools that didn't expect models of this size. And that current clouds are layering abstractions on top, that were not written with large models inference in mind.

So from now on, I'm adding "Mojo🔥Dev" to my resume, with 10 years of experience (as it's a python superset 😛). Incredibly exciting to see if these claims shake out, and we'll get the Typescript equivalent of Python but with 32K performance boost and unified inference infra