🚛💥This is how the attack on the Russian military airfield in Belaya, Irkutsk region, 4400 km from the front, took place.
One after another, FPV drones flew out from a cargo truck parked not far from the airbase. After the end of the mission the cargo container self-destructed.
Trump to Zelenskyy: “You don’t have the cards.”
Meanwhile Zelenskyy:
They combined ideas from Google DeepMind's AlphaEvolve and the Darwin Gödel Machine, trying to create a self-referential system improving itself even at the lowest algorithmic levels at *scale*.
Paper: Darwin Godel Machine: Open-Ended Evolution of Self-Improving Agents ( arxiv.org/abs/2505.22954 )
I'm so happy that Warren Buffett can finally afford to retire; it gives us all hope.
This site loads a malware popup. Avoid
ProPublica editor’s note: Out of a fear for their safety following the United Healthcare CEO killing, insurance companies asked reporters not to name staff doctors who have been accused in court of recklessly denying potentially life-saving treatments.
I NEED to go back in time and say this to a Victorian child
this is legitimately a big deal. zillow now putting climate risk on their website.
More creative humans generate more creative output from AI image generators, as judged by other people.
The effects are highly significant, but relatively modest, suggesting that human creativity matters in getting creative work from AI... but perhaps not as might be expected.
The West in coordination with Ukraine need to immediately get a grain deal in place with the new Syrian government.
This removes Russia’s only real leverage for keeping its Mediterranean presence. There will never be a cheaper, easier and safe way to severely undermine Russia.
Very excited to see the @deep-mind.bsky.social GenCast paper in press at Nature! Great work by the core team at GDM; GenCast provides a significant step forward with the ability to draw realistic sample ensemble members from the posterior forecast distribution. www.nature.com/articles/s41...
In the spirit of togetherness, if you live outside the US please share your country's most dangerous cooking method.
For no particular reason, I really like these starter packs:
Google DeepMind: go.bsky.app/GZ4hZzu
Theoretical CS: go.bsky.app/6A6GRSi
ML Theory: go.bsky.app/21nFz12
Differential Privacy: go.bsky.app/A7ABG83
A feature/bug of BlueSky is that you can back-date posts. People are using this to repost their tweets.
If you search for posts from the 1st century, there are jokes about the resurrection of Jesus. And, perhaps even funnier, some posts dated 24-11-20 instead of 2024-11-20.
bsky.app/search?q=unt...
the reason this guy runs the internet is basically nobody else is even capable of doing "your shoes were so fugly I called the FBI" and have it be a factual statement
I’ll be honest I miss the ml anons here, where’s the spark and random obsession with “Whale” or extolling the virtues of Qwen.
Maybe there’s anons in China who are obsessed with Ai2, would be something.
Made-up tech role #6:
Chief Innovation Officer of Legacy Systems.
I'm glad BlueSky is growing fast.
This was the Twitter alternative I liked the most because of the tech and its open source nature.
Twitter/X was more open in the early days and that's in part what made it successful.
It's a shame that the current X leadership doesn't understand that.
Book outline
Over the past decade, embeddings — numerical representations of machine learning features used as input to deep learning models — have become a foundational data structure in industrial machine learning systems. TF-IDF, PCA, and one-hot encoding have always been key tools in machine learning systems as ways to compress and make sense of large amounts of textual data. However, traditional approaches were limited in the amount of context they could reason about with increasing amounts of data. As the volume, velocity, and variety of data captured by modern applications has exploded, creating approaches specifically tailored to scale has become increasingly important. Google’s Word2Vec paper made an important step in moving from simple statistical representations to semantic meaning of words. The subsequent rise of the Transformer architecture and transfer learning, as well as the latest surge in generative methods has enabled the growth of embeddings as a foundational machine learning data structure. This survey paper aims to provide a deep dive into what embeddings are, their history, and usage patterns in industry.
Cover image
Just realized BlueSky allows sharing valuable stuff cause it doesn't punish links. 🤩
Let's start with "What are embeddings" by @vickiboykis.com
The book is a great summary of embeddings, from history to modern approaches.
The best part: it's free.
Link: vickiboykis.com/what_are_emb...
I’m impressed with the app and feed engineering team here @bsky.app
The UX is refreshingly crisp 🙌
By demand, I've created the final starter pack in my ML Personality Starter Pack Series.
I'm uncertain who belongs in this starter pack and so if you think you better fit in the Grumpy ML or Unreasonably Upbeat ML starter packs, let me know.
(Self) nominations welcome
go.bsky.app/5Suyk58
The weekends are for building LLMs.
Join us at reading “Build a Large Language Model (from scratch)” by @sebastianraschka.com
AI can help learning... when it isn't a crutch.
There are now multiple controlled experiments showing that students who use AI to get answers to problems hurts learning (even though they think they are learning), but that students who use well-promoted LLMs as a tutor perform better on tests.
Organizations (including university labs) should stockpile problems they could use help on from the new set of agentic & o1-like systems that are coming soon.
I suspect that they will be less useful to non-experts at first, but open up some new opportunities for complex work that AIs can do.
Easy to get the wrong impression around here, but when you actually survey students, teachers, and parents they love AI.
In the survey, it is people who never used it who don’t like it. www.waltonfamilyfoundation.org/learning/the...
[leverage voice] i'm going to infiltrate Google's internal tinder
Bluesky's firehose is a treasure trove of public data for researchers and developers, and it's completely free. Check out our developer docs: docs.bsky.app
Built my first experiment on top of Bluesky's API (actually the Jetstream WebSocket proxy) - it took ~15s of prompting in Claude to get this working: https://tools.simonwillison.net/bluesky-firehose
More details including the prompt transcript here […]
[Original post on fedi.simonwillison.net]
