@vicki@jawns.club
@vicki@jawns.club avatar

vicki

@vicki@jawns.club

#Recsys. #MachineLearning. #Engineering and distributed systems. Python, the JVM, and two small children. Writing. Terrible puns. Nutella. Building https://viberary.pizza 📚🍕

This profile is from a federated server and may be incomplete. Browse more on the original instance.

vicki, to random
@vicki@jawns.club avatar

new post on new redis ❤️‍🩹

https://vickiboykis.com/2024/04/16/redis-is-forked/

vicki, to random
@vicki@jawns.club avatar

There is this widely-held belief that current models are magic, instead of carefully-calibrated systems of components:

  1. Millions of people generating training data
  2. People aligning models through supervised RHLF / constitutional ai
  3. Human filters for model outputs
  4. People building human-centric UIs
vicki, to random
@vicki@jawns.club avatar

Something I've been wondering about it, what was different about the invention of the printing press, which also allowed potentially multiple copies of low-quality content, and LLMs? I'd love to read some historical context around what people did to combat low-quality copies in previous eras and how it worked or didn't work.

https://www.wheresyoured.at/are-we-watching-the-internet-die/

MozillaAI, to ai
@MozillaAI@mastodon.social avatar

After a year of breakneck innovation and amidst the neverending hype, how do we know if a model is any "good"?

We're excited to share our team’s learnings written by @vicki at @MozillaAI

https://blog.mozilla.ai/exploring-llm-evaluation-at-scale-with-the-neurips-large-language-model-efficiency-challenge/

Given how complex the architectures of these models are, it is crucial that the community start seriously addressing the minefield.

vicki,
@vicki@jawns.club avatar

@kellogh @MozillaAI @vicki yup sorry this is the only kind of content I produce 😂😂😂

vicki, to machinelearning
@vicki@jawns.club avatar

New post: I’ve been meaning to write something around what has fundamentally changed around the process of putting ML into prod now that we have LLMs.

TL;DR: It's still just compression, we just don't control as much anymore.

https://vickiboykis.com/2024/01/15/whats-new-with-ml-in-production/

Edent, to random
@Edent@mastodon.social avatar

Does anyone have a good primer on why Telegram seems to be the default social media platform for Russia?

vicki,
@vicki@jawns.club avatar
vicki, to python
@vicki@jawns.club avatar

Went down a rabbithole around why we use the TYPE_CHECKING constant in .

https://vickiboykis.com/2023/12/11/why-if-type_checking/

vicki, to random
@vicki@jawns.club avatar

I’m never complaining about CI that takes a minute again. https://news.ycombinator.com/item?id=38170845

vicki, to random
@vicki@jawns.club avatar

Joined a bunch of Discords to catch up on the LLM space. The energy of building is amazing but absolutely all of the information and ghost knowledge (https://vickiboykis.com/2021/03/26/the-ghosts-in-the-data/) is getting lost.

vicki,
@vicki@jawns.club avatar

@hrbrmstr @astrojuanlu yeah it’s not so much saying “Discord sucks for community”, it’s a wonderful tool for bringing people together. It’s “Discord sucks for discovery” which is important for community continuity

kellogh, to LLMs
@kellogh@hachyderm.io avatar

i wish i knew more about comparing #embeddings. anyone have resources? one thing i’ve wondered is how to convert an embedding from a “point” to an “area” or “volume”. e.g. an embedding of a 5 paragraph essay will occupy a single point in embedding space, but if you broke it down (e.g. by paragraph), there would be several points and the whole would presumably be at the center. is there a way to trace the full space a text occupies in #embedding space? #LLMs #LLM #AI #NLP

vicki,
@vicki@jawns.club avatar

@kellogh what are you looking to solve ultimately?

vicki,
@vicki@jawns.club avatar

@kellogh so somewhere between sentiment analysis and topic detection at scale?

vicki,
@vicki@jawns.club avatar

@kellogh it sound a bit what you’re talking about is averaging embeddings or coming up with unit embeddings that you can then do math on but that collapses like you said, depending on the context. There might be a different way to formulate your problem.

vicki, to Java
@vicki@jawns.club avatar

Does anyone have any recommended reading about the context behind Java’s virtual threads, the system architecture behind them and the related systems concepts?

vicki, to random
@vicki@jawns.club avatar

An extremely low-effort but high-impact thing you can do in documentation, emails, PRs, etc, is to add a sentence at the top that is just "context," giving the context of the code you're working with links to the project, whatever. Not everyone coming into the doc knows what it's about.

vicki, to random
@vicki@jawns.club avatar

What a banger of a post about what it means to care about software quality and thinking about it in a larger frame.

“When you work day in and day out with something where Quality isn't important, you take psychic damage.”

https://ludic.mataroa.blog/blog/reverence-for-the-craft-business-and-your-immortal-soul/

vicki, to random
@vicki@jawns.club avatar

The biggest insult I’ve seen software developers I respect say about software that they don’t think is good is that it is “not elegant.” Somehow hits much harder than “this is crap.”

amcasari, to random
@amcasari@hachyderm.io avatar

My alarm goes off when it's still dark, so that my brain can wake up a bit before the morning family routines. Sometimes I open Mastodon, just to glance through for any big news to start the day.

This morning, I was immediately riveted in place by @vicki's keynote on "Build and keep your context window."

Vicki expertly weaves together multiple concepts across literature, social science, history, engineering, and familiar concepts for "data people".

https://jawns.club/@vicki/111062827076966684

vicki,
@vicki@jawns.club avatar

@amcasari ❤️❤️❤️

vicki, to machinelearning
@vicki@jawns.club avatar

Enormous thank you to PyData Amsterdam for inviting me to keynote. Slides and notes from my talk, "Build and keep your context window" are all here, and if you're in Amsterdam say hi. :)

https://vickiboykis.com/2023/09/13/build-and-keep-your-context-window/

vicki, to random
@vicki@jawns.club avatar

It’s AI if you need money for GPUs, it’s ML if you need money for K8s.

dabeaz, to random
@dabeaz@mastodon.social avatar

As Python continues its ascendance, I sometimes wonder if I even have a place in that world. I'm not so sure.

vicki,
@vicki@jawns.club avatar

@dabeaz not true, as the ecosystem continues to grow and experience eternal September and draw people new to both Python and programming as a skill, we need experienced voices of reason more than ever to give context.

webology, to random
@webology@mastodon.social avatar

🏕️ "Leave it better than when you found it."

🤔 I am sure it will take me a while to update my various email and social media footers, but my ~5-year term as @ThePSF director and vice chair ended today.

🙏 It was an honor to serve, work with everyone, and leave a small mark on the community and our direction.

👟 If you are a runner and have ever finished a race a little faster than you thought you could or met a new distance goal and surprised yourself, that's how I feel today. ❤️

vicki,
@vicki@jawns.club avatar

@webology @ThePSF thank you so much for serving! 🙏🐍

vicki, to random
@vicki@jawns.club avatar

Starting a reading list for only normcore content around LLMs. What would you add?

Normcore = no hype, no langchain, no AI is going to destroy us all, but practical, technical readings in navigating in this brave new world.

https://gist.github.com/veekaybee/be375ab33085102f9027853128dc5f0e

vicki, to random
@vicki@jawns.club avatar

So funny we tried to outsource the data management layer by using pre-trained LLMs that scraped the entire internet but now we are also doing RAG so that not only do we have to manage remote API latency, but we now also still have to do in-house data management.

vicki, to random
@vicki@jawns.club avatar

📚 Pretty nervous about this, but here goes. Here's a side project I've been working on for the past few months. It's called Viberary and it's a semantic search engine. It gives you book recommendations based on ✨vibes. ✨ You enter a search query like "funny scifi" and it returns a list of (hopefully!) good recommendations.

https://viberary.pizza/

There is an about page that explains the data, model, etc. It's still pretty early stages but it's been a labor of love for me.

  • All
  • Subscribed
  • Moderated
  • Favorites
  • bokunoheroacademia
  • modclub
  • DreamBathrooms
  • InstantRegret
  • magazineikmin
  • cubers
  • Youngstown
  • slotface
  • thenastyranch
  • Durango
  • ethstaker
  • everett
  • kavyap
  • rosin
  • relationshipadvice
  • mdbf
  • khanakhh
  • osvaldo12
  • cisconetworking
  • normalnudes
  • GTA5RPClips
  • Leos
  • anitta
  • tacticalgear
  • tester
  • HellsKitchen
  • lostlight
  • sketchdaily
  • All magazines