So it's really Jeff leaving which created an opportunity for Demis to move into a scientist leadership role again?

mr. TIM@timkellogg.me · 7h ago

Jeff and Sanjay, eternal best buds

They're after every little bit of offline data that exists.

The way ai is changing software engineering you still test and evaluate engineers/researchers coding ability during hiring when in fact you don't want them to code in their day to day. Coding ability is getting worse not better, it's a bit hard to see this converging to anything positive long term?

My inbox and text message app are full of messages to myself, for just this reason — stray thoughts, musings, ruminations while running, whatever, are raw material for writing. I’d add: photos of whiteboards filled with plans, diagrams, pithy phrases, and outlines. This. Is. Writing.

Jacqueline Antonovich@jackiantonovich.bsky.social · 4d ago

*Bonus*: always remind yourself that thinking is part of writing. Actual typing of text is just a sliver of writing. So, shower thoughts, walking thoughts, random thoughts in the day about your project *is* writing. Text yourself those thoughts (or get a dictation app)

Everybody in this thread is having some sort of psychosis. When we say Americans make better coffee than Europeans I ask where in the US? Starbucks coffee is the dominant paradigm in America. There is no way.

rich traditions@helldude.bsky.social · last wk.

france, italy, constantly. enough is enough, basta even

I wouldn't mind moving to California. Since visiting San Diego, I don't see Texas the same way anymore 🤷🏿‍♂️. I almost feel like Americans downplay how great california is? I loved moving to Houston. That said, it's just heat and humidity waiting for you outside whereas CA offers so much more...

Our team just shipped Fugu-Ultra v1.1! 🐡 By dynamically orchestrating the latest frontier models, we pushed performance up by 7.9 points. We are now beating Fable 5 in complex coding and reasoning tasks without even having Fable 5 in our agent pool. Collective intelligence is the future.

Sakana AI@sakanaai.bsky.social · 2w ago

Announcing Fugu-Ultra v1.1 🐡 We’ve been thrilled by the reception to the Fugu model family. Thanks to everyone who tried it, shared feedback, and trusted Fugu with real work. Today, we’re releasing Fugu-Ultra v1.1 → sakana.ai/fugu Upgraded to incorporate the latest frontier models.

This week, we’re introducing Quests, an interactive research and exploration tool for the Atmosphere. You can ask Attie to pull information about trending topics, identify influential accounts in a particular field, or create a personalized daily briefing.

Introducing Quests

We're introducing Quests in Attie, an interactive research and exploration tool for the Atmosphere, as well as a redesign.

theliquidfrontier.leaflet.pub

1. We—@eduede.bsky.social, @mjcrockett.bsky.social, Kevin Gross, and I—have a new preprint on the arXiv today, based on ideas that emerged during an @sfiscience.bsky.social workshop in November 2024: The unintended consequences of large language models as a labor-augmenting technology in science.

The unintended consequences of large language models as a labor-augmenting technology in science

As a labor-augmenting technology, large language models (LLMs) have the potential to accelerate scientific activity across the research pipeline. But even if LLMs perform on par with human experts at ...

arxiv.org

Just got out of the odyssey and I am confused by Odysseus and his men who clearly are unfamiliar with the concept of scouting which only requires one or maybe two individuals on unfamiliar terrain not an entire garrison.

This might be the worst world cup in a long time for many reasons.

I work at Google DeepMind. This won't make me popular. But it's all public reporting: 2014: DeepMind reportedly sold to Google on conditions: no military use, independent oversight 2026: a Pentagon contract for "any lawful government purpose" Not one safeguard survived intact

anybody on here who uses multiple ai chat bots: which of them is more agreeable/disagreeable? I am spending more time using claude (past 2-3 weeks) and I find, at least in chats I've had with it, it is not overly focused on sounding agreeable at the cost of fixing my (il)logical reasoning/mistakes.

No but society needs people like this.

Dexerto@dexerto.bsky.social · 4w ago

A Norway fan has gone viral for not taking part in their Row celebration after matches and won't do it if they win the World Cup "It’s factually wrong; they didn’t row, they sailed over the Atlantic”

Correlation isn’t causation but…

This is quite the finding for a device thats just a plagiarism machine, or whatever we are calling it this week. www.nature.com/articles/s41...

Large language models can predict the results of social science experiments - Nature

Large language models can be used to estimate the results of social science experiments about as accurately as a group of human forecasters—even for experiments published after the...

nature.com

i like how do your own research turned into grok is this true

A Brown economics professor gave his students a take-home midterm this spring after students expressed anxiety following a school shooting. The majority of the students got near-perfect or perfect scores. He decided to have finals in-person; 18 dropped the class, 7 skipped the finals, the rest...

I have enjoyed the spirit of many americans on their national football team performance during the world cup who are genuinely appalled, despite the obvious difference in quality, by the way they played against Belgium. Americans are super competitive at everything: football should be no exception.

Eventually users should be able to locally host frozen version of LLMs. I don't know how you avoid losing your mind suspecting model updates of performance degradation without proper evaluation methods (also why do you even have to do that to begin with?).

Sneak peek: I implemented a 3 layer MLP that predicts the fashion MNIST dataset using SQL (with my own autograd impl inside the database). Apparently einsum notation is better expressed as a JOIN. github.com/xqlsystems/x...

xarray-sql/benchmarks/nn.py at claude/xarray-sql-mnist-demo · xqlsystems/xarray-sql

An experiment to query Xarray datasets with SQL. Contribute to xqlsystems/xarray-sql development by creating an account on GitHub.

github.com

Al Merose@al.merose.com · last mo.

It turns out SQL is an even simpler for writing NNs than JAX. With this arrays-are-pivoted-tables notion, then vmap completely goes away. You just have `grad’, `jvp` and `vjp` on columns which are Nd arrays. “Hard” tensor math is a dumb JOIN. matmul is a JOIN and GROUP BY and SUM.

asset management firms should send researchers/engineers who are good at public speaking to present at research conferences. it goes without saying. #icml26

America, the Prologue to What’s Possible. Whitman’s America. D-day’s America. America on the Moon. America of the Founding Fathers. Immigrants’ America. Americans’ America. To all about to embark on a quest: the time is now. The place is America. 3/

adding this to my reading list: arxiv.org/abs/2606.01979 arxiv.org/abs/1408.5585

It turns out SQL is an even simpler for writing NNs than JAX. With this arrays-are-pivoted-tables notion, then vmap completely goes away. You just have `grad’, `jvp` and `vjp` on columns which are Nd arrays. “Hard” tensor math is a dumb JOIN. matmul is a JOIN and GROUP BY and SUM.

Al Merose@al.merose.com · last mo.

Not yet merged, but I added autograd to my array database experiment. I’m building a new NN framework primarily in SQL. github.com/xqlsystems/x...

evergreen

Hanbin Lee@epigenci.bsky.social · last mo.

I was an idiot. XGBoost is just amazing.

Locked out of bsky on the browser because of some age verification pop up probably due to living in Texas but the mobile app seems to work, for now. Age verification is dumb, why are we doing this again?

very exciting world cup so far /s

ICML virtual pass is only $50 for PhD students.

I need a concept like convex hull, but that isn't convex. I'm trying to characterize the birds with the most extreme colors. Here I've taken 5 degree angular samples around white, and picked the furthest point within them, but you can see the big gap in the reds.