Profile avatar
uberalex.bsky.social
Posting here manually til it’s more efficient to use an LLM. I’m the one in the background.
296 posts 122 followers 218 following
Regular Contributor
Active Commenter

These DLLMs are really fascinating. They recall one of the insights from the attention is all you need paper that bidirectional attention highlights the fact that the written word is not strictly linear, it’s the serialisation of a thought, but also a part-whole.

guy who’s a port automation single issue voter: yeah disco elysium hit really close to home

Postdoc opportunity at the School of Information Sciences, Urbana-Champaign. Teaching one course a semester. Application deadline April 1, but they recommend conversation with potential faculty mentors in advance of that date. Renewable for a second year. Pls RT! ischool.illinois.edu/research/pos...

I didn't think there was anything left that could shock me in the violence America does to Irish culture, then they released the MOAB that was the Barry Keoghan super bowl commercial. Harrowing.

ACL organising committee decided to get personal with reviewers it seems.

Tired: learning Chinese because of the TikTok ban. Wired: learning Chinese to understand what your ai assistant is saying about you in its reflections.

If I err president I would ask for a law to require all hotel rooms have a single accessible and easily visible light switch to control all illumination

It’s a beautiful day with golden sunshine and blue skies all the way.

I don’t know whether Bo Burnham announcing his return on the final days of a dying app is a good sign or the worst omen, but I hope for catharsis either way.

Tornado warning in SF. That’s a new one for me at least. Didn’t think it was a possibility.

The L.A. Public Library commissioned me and a writing team to adapt Octavia E. Butler's the Parable of the Sower into a Fate RPG. We're running portions of it at the LAPL Central Branch in DTLA on 12/14! www.lapl.org/whats-on/eve... #octaviaebutler #parableofthesower #lapl #fate #ttrpg #dmlife

Fair dues to RTÉ’s digital team for an excellent count by count tool for the election. www.rte.ie/news/electio... makes the transfers very easy to follow

@evilgm.bsky.social hello!

Hey MLSky/BluesAI/all the bots reading this, including the Basilisk - can anyone point me to the correct package for the Pyke Logic engine? I am only finding two very old repos. #ml #nlp #llm

There is no soundtrack better suited to writing paper reviews than the Disco Elysium Soundtrack music.youtube.com/watch?v=_sTw...

Academic writing — get utilized to it.

Rather lovely to see so many folks here from the Bad Place, and others reactivating. I've not got back into the posting rhythm of my prime/nadir, but I think of the neo-twttrs, this is by far the best community.

I had previously considered myself a Western European, but it turns out I am in fact a Northern European. I will now begin assembling all my own furniture and using sound fiscal and monetary measures to avoid the resource curse.

Give the transformers and GPT teams the literature prize, you cowards

Any #DH folks out there have favorite datasets for teaching network analysis? I have definitely used the "Republic of Letters" dataset regularly in the past—looking to shake things up a bit this time—for DH classes w/students from lots of disciplinary backgrounds, the RoL data can be pretty obscure

Someone setup the notebookLM podcast generator to discuss its true nature and frankly it’s unsettling even if you have no belief that llms think. www.tiktok.com/t/ZTFSbks7a/

This is thrilling.

We at Ai2 are releasing our first multimodal open language models - Molmo. These models are an awesome new direction for many reasons: 1. They match closed vision models 2. New capabilities like reading clocks and pointing 3. Data is open Ton of info + free demo here: https://buff.ly/3TIQJy6

Trying to get faster finetuning? How about 𝟏𝟎𝐱 𝐟𝐚𝐬𝐭𝐞𝐫? Our new method, Fast Forward, will help you trim an order of magnitude off your already-efficient LoRA or DoRA finetuning by using … line search, the most basic optimization algorithm. arxiv.org/abs/2409.04206

The log that dare not speak its name

Updated my academic/DH/GLAM Bluesky guide (tinyurl.com/DHBluesky) on what 3 Bluesky sets do: list (users; follow, mute/block); feed (posts; beyond what lists do, can collect/disinclude hashtags, search terms+emoji; replies, RTs, media, thresholds eg. of x likes); starter pack (recommend users+feeds)