In which I discuss some books I read last year magnusross.github.io/posts/2025-b...

It’s a big day for central banks around the world today, so naturally Alphaville is tackling the big topics — like whether Rishi Sunak got some cowboy boots as PM and how much they cost. www.ft.com/content/82d2...

Could revealing Rishi Sunak’s cowboy boots ruin Brex-Tex relations?

There’s a Sunak in my boots!

ft.com

🚨 New paper alert 🚨 Our latest work on using machine learning to help clinicians to switch patients from IV to oral antibiotics more efficiently: openreview.net/pdf?id=j8wsq...

I wrote about a crazy new LLM benchmark where the models are let loose on the crypto markets. Spoiler: sticking all in the prompt does not work that well in this case! magnusross.github.io/posts/nof1-a...

I wrote something about building systems people actually want as an academic in ML. It's pretty much an open letter to 6-months-ago me. magnusross.github.io/posts/moms-m...

The American guns turning Haiti’s gangs into an army https://on.ft.com/4lVGHFY

The American guns turning Haiti’s gangs into an army

Weapons smuggled from Florida are empowering the country’s militias and enabling them to challenge a fragile government

on.ft.com

Come to our seminar with @wessel.ai on "Foundation Models for the Earth System", it's going to be a cracker! It's open to non-UCL people also.

Good to get some recognition for my incessant posting of Ben's blogs in the Alphaville comments section!

Ben Recht@beenwrekt.bsky.social · last yr.

Testing the waters of finance blogging, I wrote about deep learning memes, random features, and the glory of kernel smoothing.

Take note of the legal advice from Amnesty International and more detailed information from @netpol.org below. CND supporters should be aware that point 4 could impact on how supporters refer to non-violent direct action.

ChatGPT is down but The Museum of English Rural Life still stands, proving once again that Silicon Valley cannot compete with the history of rural England and its people.

What news editors and politicians think people want: "Just tell them they'll get a tram somewhere in their city if they're nice." What people really want: "SHOW ME THE FULL GPS ROUTE PLAN COMPLETE WITH LINKS TO THE WIKIPEDIA PAGES OF THE DISUSED RAILWAY LINES THAT COULD COME BACK INTO USE."

Cancer Research UK today published a report which highlights changes in cancer survival rates over the last 50 years. Many headlines (like this one from the Guardian) have chosen to report the ‘doubling of cancer survival rates’ since the 1970s. But that's misleading... 1/18

A new paper just dropped from Tri Dao(🐐)'s lab! arxiv.org/abs/2505.21487 Here is my hot take!

Hardware-Efficient Attention for Fast Decoding

LLM decoding is bottlenecked for large batches and long contexts by loading the key-value (KV) cache from high-bandwidth memory, which inflates per-token latency, while the sequential nature of decodi...

arxiv.org

the inexplicable core of Starmerism, as a truly new formation in world politics, is seeing governance as wholly an extension of general election campaign messaging

George Pearkes@peark.es · last yr.

this is a very strange message from Starmer. First, you're boosting the far right which is, uh, risky. Second, your government was elected LAST YEAR why are you talking like the election campaign is under way?!?!! *STARMER: TORIES SLIDING INTO ABYSS, CHOICE IS LABOUR OR REFORM

Big news!! 🚀 Aurora, a foundation model for the Earth system, has been published in @nature.com. A massive congrats to the whole team! Very proud of this achievement, and thrilled to see that it’s finally out there. 😊 www.nature.com/articles/s41...

A foundation model for the Earth system - Nature

Aurora, a new large-scale foundation model trained on more than one million hours of diverse geophysical data, outperforms operational forecasts in predicting air quality, ocean wave dynamics, tropica...

nature.com

Short write-up about how I got xAI to briefly add "white genocide" stuff back into Grok's system prompt. smol.news/p/the-utter-...

The Utter Flimsiness of xAI’s Processes

xAI's was happy to put "white genocide" stuff back into Grok's system prompt without second thought.

smol.news

it's very odd watching the election news in the UK when Labour still has *years* left on their massive governing majority they can just ignore this stuff and create prosperity and get popular and what's particularly weird is that it's not just commentators it's Labour themselves that don't get it

You know you're in for a banger paper when the this is the first line of the abstract...

Stealing from X for clout, but someone bought a staggering 25% of the *entire* open interest of a semiconductor ETF (which is 20% NVDA) in a single OTM put option yesterday and today. $3.7 billion of underlying stock, 8ish delta, $23 mil of premium, so levered 15:1 betting on the NVDA embargo news.

one for the debtors!

Ollie Clark@oc-econ.bsky.social · last yr.

I recently reviewed The Paradox of Debt by Richard Vague for the Society of Professional Economists. Check it out at the links below ⤵️

A new Python edition of "Forecasting: Principles and Practice" is now available online at otexts.com/fpppy/. Thanks to @azulgarza.bsky.social, Cristian Challu, Max Mergenthaler, Kin Olivares & Nixtla for making this happen. #forecasting #python

I just came across this paper from ICLR 2024 which proposes an intricate combination of transformers and diffusion models to generate forecasts with these uncertainty bounds (red box), which are clearly inappropriate and could likely be outperformed by modelling the data as a random walk...

CUDA mode -> Triton mode

Paul Chang@mummitrollet.bsky.social · last yr.

I've been diving into the "black magic" world of CUDA recently. More posts may follow, but I think we're at an interesting point. Perhaps the "CUDA moat" is under pressure and perhaps changing how we interact with GPU programming. 🧵

13. Third, that because this is a new _social_ technology, its uses involve human relationships mediated through technology. When someone uses ChatGPT to craft a resume, they are engaged via an LLM in a social relationship with all the other humans who wrote the resumes etc in training set ...

This blog on the North Korean Bybit hack — what they reckon is the largest heist of all time — is pretty cool www.elliptic.co/blog/bybit-h...

Multi-Head Latent Attention vs Group Query Attention: We break down why MLA is a more expressive memory compression technique AND why naive implementations can backfire. Check it out!

@datacrunch.io · last yr.

⚡️Multi-Head Latent Attention is one of the key innovations that enabled @deepseek_ai's V3 and the subsequent R1 model. ⏭️ Join us as we continue our series into efficient AI inference, covering both theoretical insights and practical implementation: 🔗 datacrunch.io/blog/deepsee...

1/ Introducing ACE (Amortized Conditioning Engine)! Our new AISTATS 2025 paper presents a transformer framework that unifies tasks from image completion to BayesOpt & simulator-based inference under *one* probabilistic conditioning approach. It's Bayes all the way down!

auf 'stack, I return to regional policy and (piano dischord) economic geography. open.substack.com/pub/backofmi...

ramsay's regional policy nightmares (slight return)

but reykjavik! but dallas! but tilbury!

open.substack.com