Vicki

ATProto makes programming fun again. Here's a side project I did over the holidays, a small app that surfaces the 10 latest links to GitHub repos from the network. I learned a ton, including Go, Javascript, and a lot about websockets vickiboykis.com/2025/01/23/y...

The first time I listened to Berghain by Rosalía I understood that I was witnessing new, true art.

Shower thought

feels like I’m the only one pacing tbh and that’s only because I’m getting docker build errors

Startup idea: small personalized models that are fine tuned on the trauma of your past production incidents

Area woman continuously working in recsys more and more convinced that most ML problems are recsys problems

It’s pretty wild how cleanly the labs have been able to spin basic security incompetence into positive marketing

New post: we need to be deleting more code. vickiboykis.com/2026/09/01/b...

Blog posts we used to have: “how I reduced latency across several hundred machines by 200 ms by removing two O(n)^2 hot paths” Blog posts we have now: “Here’s how to get the most out of /ultrathink in the latest release of Claudex 3.0.5”

Curious about people’s ML dependency management strategies in the face of supply chain risks. Often pinning dependencies particularly in the torch/HuggingFace ecosystem is the only way to create reproducible training and inference code but that means dependabot isn’t happy.

Eye bleach: an actual architecture and engineering blog post lobste.rs/s/ko1ji1/lob...

Looking to read an in-depth examination of the project history and code evolution of Bun in Zig, along with a technical evaluation of its current Rust re-implementation. I want to check out links to PRs, sidebars on Zig and Rust, the works. Who’s doing this?

Recommended reading if you feel you’ve lost your way , or if you’re having more fun than ever

traces are just logs right

This is the only thing noam shazeer has blogged about publicly and I use it all the time medium.com/@NoamShazeer...

new post: how I develop recently using local models. the tooling is now good enough to do agentic workflows and everyone should give them a try! vickiboykis.com/2026/06/15/r...

The machine learning circle of life

every day we get closer to looping back around to small purpose-built specialized models

Running a local model and harness (Pi + GGUF via LM Studio) and I can see everything it's doing, which is dope.

noodling on something

Good post from the author of rsync on the recent pile-on: medium.com/@tridge60/rs...

It’s not “do good work and be successful.” It’s doing good work, talking about the work, figuring out how to get your work to the right people, accepting that good work is slower than bad work and finding a path through that anyways

Great piece, “We're entering a world where it's cheap to produce bigger changes, but that doesn't make it any safer.” apenwarr.ca/log/20260531

Very strongly recommend breaking out of the token discourse and touching grass from time to time to maintain a grip on reality

New post: we should be more tired than the model vickiboykis.com/2026/05/28/w...

doing an online search and coming across a quora answer that is not only written by a real human but by Adam D'Angelo is a jump scare coming from the dead internet www.quora.com/How-can-I-ge...

i miss simclusters so much

Legitimately feels like an unquantifiable vibe shift the last few weeks where the pendulum is swinging back to reasonable takes and people experimenting with model choice 🙏

Anyone running a bot that summarizes their discords over a given time? Is this something that’s possible with the API?

I find that I can’t concretely commit to an implementation until I touch the code a bit and one thing that it has gotten easier than ever to do is to get to a draft PR where you share your proposed approach and easier see the trade offs of certain design decisions

this has to be the paper that, looking back at the authors from today's context, you're like, oh damn all those people really did work together arxiv.org/abs/2005.14165

this tab ordering tells me everything I need to know about GitHub’s actual product priorities for the next year

Yes, I think my current work on the Hardwood Parquet file parser is a good example: while I can implement it without a lot of dependencies (I decided to ship my own AWS API request signer), hundreds of hours (of LLM usage) went into the project, ensuring correctness, great performance, etc.

I don’t buy the idea that libraries will go away since everyone can just write their own now. Maybe for small internal company use-cases. But libraries are not just code, they’re time invested in edge cases, they’re hundreds of hours of human eyes. I don’t want to replace that.

Just once I wanna see a research paper say, the tokens we pretrained on were not high-quality or diverse, they were absolute trash we scraped from AMA subreddit longtail submissions combined with everyone's public Venmo transaction messages

DeepSeek v4 Flash with *local inference* after 24h of playing with that: even with the 2 bit selective quantization GGUF, iti is the FIRST time I feel I have a frontier model running on my computer. This is *crazy*, and probably a much stronger change in the landscape than PRO.