one small step for figma, one giant leap for design systems

Shriram Krishnamurthi 🟤 🏴‍☠️ 👨🏽‍🏫 🚴‍♂️ 🏏 Micah Cantor I’m not sure what a paradigm is or how to define it, I can only point at its shadow. Boxing general purpose programming languages into a single paradigm or multi-paradigm feels reductive…

I agree with this take, but duplicate code is often better than a very bad abstraction.

Introducing the Little Languages Reading Club, a monthly/bi-monthly club where we collect set of papers/essays on a given topic and dive in.

github.com/Little-Languag…

MLX team needs a raise.

I’m going through my list of ML updates (96 total) from the past two weeks and MLX has 7 entries I think need to be shared. This is after trimming it down to about half. Their pace is incredible.

Mark your calendars for #WWDC24 , June 10-14. It’s going to be Absolutely Incredible!

Fuchsia as Google's AIOS?

Wrote this over the weekend inspired by AICI (IMO we should talk about this more): Semantically-Aware Constrained Decoding for Code Generation

krismuniz.com/posts/constrai…

Drake's gardener uses this rake to rake the annoying yellow flakes that you get when the wind breaks the leaves in your garden. #ad

Toxicity checks should test for bad rhymes

Alex Albert Support something a bit like github.com/microsoft/aici so I can send my own logic for next token guidance (in WASM or something else) to the API for much more finely grained control over generation

github.com/microsoft/aici

This is an amazing idea. WebAssembly will low-key change the world. We'll all soon be sending tiny little programs with our API requests running in sandboxes.

This is actually brilliant work by microsoft: github.com/microsoft/aici.

Sending mini WASM programs instead of prompts to LLM providers would be incredibly powerful. Multiturn, , Grammar, COT, RAG, function calling, etc could all massively benefit from this

MLX is amazing.

Dare Obasanjo🐀 'Just develop a scoring metric so that the system itself can tell whether one prompt is better than another, and then just let the model optimize itself'

The word 'just' is doing a whole lot of work there!