Refreshing to hear people calling the bullshit what it is in @science.org Full article www.science.org/content/blog...

The public overwhelmingly rejects OMB's federal grant rule change. I used the GPT 4o mini LLM to classify the comments, showing 94% in opposition. Of the 5% supporting, ~84% are verbatim copies. This blog post summarizes the results (full analysis linked in post). www.jhelvy.com/blog/2026-07...

The public overwhelmingly rejects OMB’s proposed federal grant rule – jhelvy.com

OMB has proposed a sweeping rewrite of the rules that govern every federal grant. Thousands of people filed comments. I downloaded them to look at what they actually say.

jhelvy.com

There are 496,769 comments on the OMB rule that will devastate U.S. science. Thank you 🙏 Little doubt this administration will shove this rule through anyway, but these comments will matter. The fight to defend science is not over. Many more battles to come. www.regulations.gov/docket/OMB-2...

In 48 hours with #Claude code, I built a C++ backend to my logitr #rstats 📦 to make mixed logit model estimation ⚡ fast. There's no excuse now for slow software now when tools like CC are available to do the performance engineering. A post on this build out: www.jhelvy.com/blog/2026-07...

Dissenting on a birthright citizenship case should be disqualifying for SCOTUS. Not only is the language of the 14th amendment abundantly clear, we fought a civil war to enshrine it in our constitution. That amendment was a very, very costly amendment. Shame on Alito, Gorsuch, and Thomas.

This is an excellent talk on the recent history of AI progress in solving increasingly challenging problems in science and the shockingly consistent outcome that just scaling the models more leads to continued performance improvements. youtu.be/Mw60FH5iflI?...

Training Sand to Think: Artificial General Intelligence & Future of Physics

YouTube video by Perimeter Institute for Theoretical Physics

youtu.be

If you've ever been curious about how to use claude code skills, check out @pingfanhu.bsky.social's recent post. He shows how to leverage skills to encapsulate complex workflows and achieve impressively consistent outputs from agents. pingfanhu.com/blog/2026-06...

Agentic Skills – Pingfan Hu

My approach to agentic skills and how I build them around my own practice.

pingfanhu.com

Excellent reporting and analysis on America's deadly mega trucks and SUVs www.nytimes.com/interactive/...

The Deadly Rise of Giant Trucks and S.U.V.s

The vehicles on American roads have grown larger — and they are killing thousands more pedestrians, a Times investigation found.

nytimes.com

🚨 The U.S. scientific enterprise faces an existential threat few are talking about. A proposed OMB rule would reshape ALL federal research funding — NIH, NSF, DOE, DOD, every agency. What it does & how to push back 👇 defendfederalresearch.org 🧵...

🚨 5 alarm fire territory 🚨 www.nytimes.com/2026/06/05/w...

Police Remove Diabetes Experts From Conference for Distributing Critique of Trump Administration

nytimes.com

I cannot think of a better fitting president for the 二百五 bill

The BUILD America 250 Act charges #EV owners $130/year to "pay their fair share" in road taxes. Hogwash. Average drivers pay ~$95/yr in federal gas tax. A 50 mpg hybrid pays just $42, yet no surcharge for them. It's almost like this has nothing to do with tax revenue and is just an EV penalty 🤔

Esp for use with LLMs - I'm increasingly convinced #quarto is the ideal LLM output. Separating *content* from *format* is so, so useful. www.jhelvy.com/blog/2026-05...

The Unreasonable Effectiveness of Quarto – jhelvy.com

Why you should start asking your favorite LLM (cough Claude cough) to produce .qmd files

jhelvy.com

Antoine Vernet@antoinevernet.com · 2mo ago

And Quarto to bring them all, and in the darkness bind them

#Quarto is much more than a publishing framework. When leveraged with agents with Claude code, it's a powerhouse tool for agentic workflows. In this blog post, I make a few quick arguments for why you should check out Quarto if you're working with agents. www.jhelvy.com/blog/2026-05...

The Unreasonable Effectiveness of Quarto – jhelvy.com

Why you should start asking your favorite LLM (cough Claude cough) to produce .qmd files

jhelvy.com

A less extreme version is just to put speed governors on every car. You can put the cap at something like 85mph, but seriously beyond that there is truly no reason to be allowed to drive faster except like running from the cops

Michael Hobbes@michaelhobbes.bsky.social · 12mo ago

My least popular (and most correct) view is that cars should be automatically limited to the local speed limit. Put the pedal to the floor and you still can't go over 25mph in a residential area. (15 in Manhattan btw)

We are watching our democracy crumble. With this SCOTUS ruling, every state has no choice but to follow TX and redraw their voting maps. The south will become an entirely white R vote, and blue states will become entirely D. The changes made now will lock in polarization for a generation or more.

Excellent piece by Geraint Rees and @jameswilsdon.bsky.social on the massive challenges facing the legacy grant funding system in the era of AI. We need a better approach, like yesterday. www.nature.com/articles/d41...

Could agentic AI topple grant-funding systems?

Funders must take action before an increase in submission of high-quality proposals written using AI models makes it impossible to sort the wheat from the chaff.

nature.com

You're next Donny www.nytimes.com/2026/04/12/w...

Orban, Beacon to the Right, Concedes Defeat in Hungary’s Election

nytimes.com

@pingfanhu.bsky.social has an amazing 3-part blog series on agentic engineering workflows: Part 1 - Claude code basics: pingfanhu.com/blog/2026-03... Part 2 - Workflow and examples: pingfanhu.com/blog/2026-04... Part 3 - Software that ties it together: pingfanhu.com/blog/2026-04...

Agentic Engineering - Part 1 – Pingfan Hu

My experiences of effectively using AI for personal workflows, with a focus on Claude Code.

pingfanhu.com

At its peak, the US had over half a million soldiers deployed to Vietnam, with a total of 2.7 million served over the course of the war, and US forces still failed to secure the country

This is monumental...the entire open source world now has access to a #python version of Claude code, with a #rust port coming soon. This is the fastest repo in history to surpass 100k stars. Absolutely wild that the most powerful AI coding agent is now just out there. github.com/ultraworkers...

GitHub - ultraworkers/claw-code: [Notice] The repo temporarily locked while ownership transfer. in the meantime we maintain on here: https://github.com/ultraworkers/claw-code-parity. The fastest repo ...

[Notice] The repo temporarily locked while ownership transfer. in the meantime we maintain on here: https://github.com/ultraworkers/claw-code-parity. The fastest repo in history to surpass 100K sta...

github.com

Oof...damning evidence of AI responses on online surveys. The battle against AI survey bots is well underway, and it doesn't look like the humans are winning www.theguardian.com/technology/2...

‘Our assumptions are broken’: how fraudulent church data revealed AI’s threat to polling

Experts say paid participants are using automated tools to generate unreliable survey responses at scale

theguardian.com

The very best stats course you can get is still free. Absolutely amazing stuff.

Richard McElreath 🐈‍⬛@rmcelreath.bsky.social · 5mo ago

Statistical Rethinking 2026 is done: 20 new lectures emphasizing logical and critical statistical workflow, from basics of probability theory to causal inference to reliable computation to sensitivity. It's all free, made just for you. Lecture list and links: github.com/rmcelreath/s...

@jigarshahdc.bsky.social on point👇

Jigar Shah@jigarshahdc.bsky.social · 5mo ago

Europe just ran one of the most important energy transition experiments in history. After the Russian invasion of Ukraine, Europe had to rapidly replace Russian gas imports. Working with 27 member countries will different languages was not easy, they got it done in years not decades. Thread 🧵

If you think that's cool, wait til yoy see my market research tech

Hypervisible @hypervisible.blacksky.app · 5mo ago

“Instead of paying humans to join focus groups and complete surveys, Aaru uses thousands of AI agents, or bots, to simulate human responses. It feeds demographic and psychographic information into its models to create human profiles that match clients’ needs…”

Has there ever been a time when the strategy of "if we just keep bombing civilians eventually they'll finally rise up to overthrow their oppressive government" ever worked?

Someone ported Karpathy's microGPT to pure #julialang in 99 lines with 0 dependencies, it's 1600x faster than CPython... hard to emphasize enough how impressive this is www.reddit.com/r/Julia/s/Vg...

From the Julia community on Reddit: I ported Karpathy's microgpt to Julia in 99 lines - no dependencies, manual backprop, ~1600× faster than CPython and ~4x faster than Rust.

Explore this post and more from the Julia community

reddit.com

@simonwillison.net is the first I've heard use the term "agentic engineering", which feels like a much better description of how I use tools like Claude Code (as opposed to "vibe coding", which implies no human architecting the result). Great stuff in this post: simonwillison.net/2026/Feb/23/...

Writing about Agentic Engineering Patterns

I’ve started a new project to collect and document Agentic Engineering Patterns—coding practices and patterns to help get the best results out of this new era of coding agent development …

simonwillison.net

Sharing a new #opensource project we've been building: vehicletrends.us — an interactive dashboard tracking U.S. automotive market trends from millions of listings. Data is available via the {vehicletrends} #rstats package. Site made with #Quarto, code at github.com/vehicletrends 🚗 📊

Check out this wide-ranging discussion I had with Frank Sesno, director of the Alliance for a Sustainable Future at GWU. Topics include EV competition, US-based EV production, facing the challenge of Chinese EV leadership, and national security (among other things). www.youtube.com/watch?v=oUMz...

At the EV Crossroads: China Surges, the U.S. Stalls - The Greenhouse

YouTube video by GW Alliance for a Sustainable Future

youtube.com