Shritesh Bhattarai

CTO @dylibso

[Jeff Lindsay @progrium]

wanix is so good now i was like "damn i gotta set up the vm and a linux environment" with older wanix in mind, but then oh, right: copy-paste 4 lines of html and boom linux

[Max Leiter @maxleiter]

"They're made out of weights."

"Weights?"

"Weights. Floating-point numbers. We checked the whole thing through. It's nothing but weights."

"Weights doing what? Where do the words come from?"

"The weights make the words. Are you understanding me?"

[Shritesh Bhattarai @shritesh]

@DNAutics @nilslice nominative determinism strikes again

[Shritesh Bhattarai @shritesh]

“everyone should be able to make their software do whatever they need, and it should just work as changes are applied, revised, and reverted across time, across multiple users, across system boundaries. AI is making this real”

[steve.prophet @nilslice]

https://t.co/HiRMVHj65k

[José Valim @josevalim]

Elixir v1.20 released! Now officially a gradually typed language: Elixir type checks every single line of code, finding bugs and dead code, without developer overhead (no typing signatures) and extremely low false positives rate. Plus a faster compiler! Links and reports below.

[Sebastian Aaltonen @SebAaltonen]

Many renderers have callbacks. Ours does not. WASM->JS callback would be horribly slow. If you want to run custom code in the rendering, you write a compute shader. Compute shader has access to all the same (interpolated) engine/camera globals as render passes. It can calculate/interpolate data and write to any UBO/SSBO you bind into your materials.

Our plan is to be a fully GPU-driven renderer. CPU doesn't even know what is visible. CPU callbacks wouldn't do much good in a modern architecture like this. Custom setup code has to be a compute shader.

[David Crawshaw @davidcrawshaw]

I understand the concern of skills atrophying when using agents. But so far I am not seeing it. Instead I have learned all sorts of dark secrets of linux networking I somehow didn’t learn before agents building a networking product.

[José Valim @josevalim]

Back in Jan 2025, I suggested to @chris_mccord that https://t.co/JgfAy72se8 should not have many bespoke tools, instead the agent should just write code (elixir/js/whatever) to achieve what it needs. Today this is pretty much a standard across agents:

* Codex uses shell (sed/rg/etc) to read files
* Claude Design canvas is pretty much HTML + React components
* Tidewave's browser control is a single tool that runs JavaScript with additional functions
* Claude Code new dynamic workflows is a single tool that runs JavaScript with additional functions

The whole MCP code mode is inverting the orchestraction from tools back to code. :)

[Shritesh Bhattarai @shritesh]

To make agents successful, we’ve built systems that are also really nice for humans.

[Gary Bernhardt @garybernhardt]

I think it's important that we continue taking a moment occasionally to remember that MACHINES CAN WRITE NOVEL CODE AND IT ACTUALLY MOSTLY WORKS lest we forget how amazing this is

[Yining Karl Li @yiningkarlli]

Here's a fun blog post about bytecode VMs in surprising places.

Mildly hot take of mine: a bytecode VM in a GPU kernel is not as bad of an idea as one might think, and in some cases it can actually be the best solution. Fun examples in thread: (1/5)

https://t.co/UjkQdjcnDE

[ruperts.world @rupertmanfredi]

we are so in the DOS era... it's time for the GUI

[Georgi Gerganov @ggerganov]

Highlighting the new WebGPU backend in llama.cpp/ggml

The work to bring full-fledged WebGPU support in llama.cpp started about an year and a half ago. It has been lead by @reeselevine and team at USCS.

For more information, checkout the interactive blog and paper in the quoted post. Here are 2 excerpts from the paper, summarizing the implemented software architecture.

[Shritesh Bhattarai @shritesh]

An Effect Notation Based on With-Clauses and Blocks - Yoshua Wuyts https://t.co/tpiKejnCH3

[lcamtuf @lcamtuf]

The cat's out of the bag! My latest book, "The Secret Life of Circuits", is available in early access:

https://t.co/ormpiPwapu

It's what I wish I had when I was starting out. Electrons to embedded systems, 290+ color illustrations and 420+ pages of well-explained theory.

[José Valim @josevalim]

We have just wrapped up the meta-issue which tracked type inference of all Elixir constructs: https://t.co/HvOu1Iw9Cv

It includes typing of the construct themselves and features like occurrence typing, to provide a high-degree of precision. Expect a new RC soon as we will continue measuring performance on our way to Elixir v1.20.

[Shritesh Bhattarai @shritesh]

cuda-oxide: The Safety Model https://t.co/xRhP5RTTAc

[Jarred Sumner @jarredsumner]

why: I am so tired of worrying about & spending lots of time fixing memory leaks and crashes and stability issues. it would be so nice if the language provided more powerful tools for preventing these things.

[Cosmos Raj @cosmos_raj]

We are recreating Nepali culture from first principles

[Jamie Turner @jamwt]

If you want happy users, you should think about system performance the way BEAM/OTP/Erlang does.

Stop wasting time on bullshit benchmarks.