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”
Vicki
@vickiboykis.com
Founding list[float] engineer. Recsys. Personalization. Infra. Systems. Normcore code. Nutella. Vectors. Words. Vibes. Bad puns (soon). https://vickiboykis.com/what_are_embeddings/
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.
There are like only three design patterns you need to use 80% to make stuff good and fast and elegant Make it a lookup (hashmap) Split it up into smaller pieces (sharding) Combine it into one large piece (map reduce)
Eye bleach: an actual architecture and engineering blog post lobste.rs/s/ko1ji1/lob...
lobste.rs is now running on SQLite
87 comments
lobste.rs
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
Everyone working on recsys/embeddings/information retrieval/search is sleeping on this newsletter if you're not subscribed. I can't believe I didn't mention it before but it just summarizes the best content in the field every week recsys.substack.com
Top Information Retrieval Papers of the Week | Sumit | Substack
A weekly curated newsletter about the latest research papers in the Information Retrieval domain, including Recommender Systems, Search, Retrieval, and Ranking. Click to read Top Information Retrieval...
recsys.substack.com
the actual most revolutionary thing the ai labs have actually done is make it look like it's insanely easy to duct tape together python microservices with all their concurrency issues and trained models with all their transformers issues and have them run as one api call
This is the only thing noam shazeer has blogged about publicly and I use it all the time medium.com/@NoamShazeer...
Shape Suffixes — Good Coding Style
If you code neural networks, I believe that this convention can make your life more pleasant. We keep this pretty religiously at…
medium.com
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...
Running local models is good now
Local agentic coding has gotten great over the past few months
vickiboykis.com
this is so cool, continuing my local experiments and using pi and lm studio and gemma to introspect how i use agents locally. On the left, you can literally see the streaming tokens and the context window filling up.
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.
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
The software industry: annealing, but wrong
In recent months I've heard of several teams with an interesting policy: each pull request should be no more than a few files, and no more t...
apenwarr.ca
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...
We should be more tired than the model
Adding deliberate friction back into development
vickiboykis.com
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...
Something happens to my brain after agentic coding that I can’t describe. It’s like cognitive offloading which folks have already written about, but even more. It feels like I can’t think through problems anymore. Like a fog. Using agentic but losing my hard-won agency.
New post: Tagging my blog with BERTopic and LLMs. I suspect as token costs increase, we'll see more blended traditional ML/LLM systems. And, in general, they work really well! vickiboykis.com/2026/05/18/t...
Tagging my blog posts with BERTopic and LLMs
LLMs mean you still need a human in the loop, but in a different part
vickiboykis.com
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
Language Models are Few-Shot Learners
Recent work has demonstrated substantial gains on many NLP tasks and benchmarks by pre-training on a large corpus of text followed by fine-tuning on a specific task. While typically task-agnostic in a...
arxiv.org
this tab ordering tells me everything I need to know about GitHub’s actual product priorities for the next year
