proof of concept safety for zig: www.youtube.com/watch?v=ZY_Z...

"The Supreme Court blocked it, but that didn’t stop me." - donald trump

highly suggested civics lesson for the era (yes, this is how it works): youtu.be/n1KgxqEQn0A?...

I'm here for dark sabine youtu.be/shFUDPqVmTg?... if douglas prasher (discoverer of GFP) could survive driving a courtesy van at a toyota dealership in huntsville, TeV physicists will find a way to survive too.

pcr skillz😌✅

vim ❌ emacs ❌ vscode ❌

For my last programming project, I built a proof-of-concept borrow checker for #Ziglang: github.com/ityonemo/clr

just finished my last github repo =( #elixirlang #ziglang

has anyone put tailscale on #elixirlang #nerves? maybe @fhunleth.bsky.social knows?

with refinement type analysis, it's possible to track units in #ziglang

"ownership" tracking in #ziglang is possible via analysis of AIR. The first line is the runtime detection and the red error is the compile-time detection

home molbio lab LFG!!

I just ordered a 744-BP synthetic gene for $144. Times have changed.

my gripe with the #elixirlang typesystem is that if you want checking against structs in your function header, you must specify the struct module, which creates an extra map lookup in the function call

basic #ziglang use after free detection

here is static analysis analyzing AIR and detecting "undefined" use in #ziglang .

It's possible to do static analysis on #ziglang AIR to detect stack pointer escapes. Should also be possible to track lifetimes of heap-allocated pointers (preventing double free and UAF), prevent "undefined" use, and do other "dependent type" things like units, etc.

FYI Roommates.com is officially running #ElixirLang in production. Go check it out!

reposting @dnautics.bsky.social predictions. LCDM looks to be about to lose the lambda to Timescapes theory

quantum AI, except the "quantum" is using complex-valued activations and weights

Don't worry, I promise to maintain zigler till 1.0

fine-tuning llms on #elixirlang is most definitely a thing: (before/after)

cookin'

I hate python so much, but it is doing the thing

attempt #1 at lora-training 💀 OOM

cooking

after playing around with vllm's function calling interface, i can safely say "it's no better than telling the llm to `respond with JSON` in the system prompt"

Taking a short side jaunt into voice activated LLMs

erlang's undefined is a worse version of elixir's nil #isaidit

I'm playing around with building an elixir-first LLM. This is an LLM that is trained on elixir code. Looking for someone to try it out for me. Should be in a "first usable state" in about 2 weeks. Any guinea pigs?

Phi-3-mini-4k-instruct 😑 ``` defmodule Fibonacci do def loop(0) do [0] end def loop(1) do [1, 1] end def loop(n) do a = 0 b = 1 fibs = [a, b] i = 2 while i < n do a = b b = a + b push fibs, b i = i + 1 end fibs end end ```

today: - f'd around with amd 125m. terrible model. - went through LORA fine tuning exercise (will take a look tomorrow) - found a few SLMs that might be good to build an #elixirlang LLM (phi-3-mini, replit-code)