iiinova — where your digital self keeps evolving

John Paul Helveston @jhelvy.bsky.social · 10/10/2026

Yeah I'm with others on the music. I think a good audio setup and no music would come off better. Human voice > AI voice, but need a good audio set up. It's the environment, not the 🎙️

030

John Paul Helveston @jhelvy.bsky.social · 06/10/2026

What blew me away was it's even in the same order

091

John Paul Helveston @jhelvy.bsky.social · 06/10/2026

Walked into this today, and all I can think about is penguins #rstats

Three urinals with different color inserts: purple, teal, and orange (which resemble the color palette from the palmerpenguins R package)

5412

John Paul Helveston @jhelvy.bsky.social · 29/09/2026

Rolling back fuel economy standards during the worst Middle East oil crisis since the last Middle East oil crisis that prompted Congress to create those standards in the first place is...a choice

030

John Paul Helveston @jhelvy.bsky.social · 24/09/2026

earmark can now run on GitHub. Set up a repo, then add articles by listing them in a sources.yml file. Making a commit kicks off an action that converts the articles into mp3s that are served on GitHub pages. Listen to your articles from any podcast app! earmark-dev.github.io/earmark/gith...

earmark-dev.github.io

Set it up – earmark

0121

John Paul Helveston @jhelvy.bsky.social · 16/09/2026

Sharing a great little 6-min video by a creative film maker and story teller. Looks like he just getting started on some really cool work. youtu.be/IXRVD7ZqncQ?...

youtu.be

Healthy things stop growing. Why can't economies?

YouTube video by Your Local Planet

001

John Paul Helveston @jhelvy.bsky.social · 13/09/2026

On my way to posit::conf(2026)! Looking forward to sharing ideas about data, code, AI, & tools with a fun and ridiculously creative community. And if you'll also be there, come to my talk Tues afternoon to see how we used #quarto and AI agents to build the vehicletrends.us dashboard!

020

John Paul Helveston @jhelvy.bsky.social · 09/09/2026

Terence Tao on OpenAI's Navier-Stokes solution

Jeff Goldblum: "Your scientists were so preoccupied with whether or not they could...they didn't stop to think if they should."

020

John Paul Helveston @jhelvy.bsky.social · 03/09/2026

Journal editors can't find reviewers because we're overwhelmed. I checked my gmail, review requests have more than doubled since the pre-LLM days. This is also why you're seeing more AI-generated reviews - reviewers can't keep up and reach for the LLM to get the job done faster (not better).

Bar chart of journal review requests per year, 2012–2026. Requests grew slowly from 3 (2012) to ~30/year (2019–2023), then jumped sharply after ChatGPT's release, hitting 86 in 2024 and 73–80 in 2025–2026.

Bar chart of journal review requests per month, 2012–2026, with a smoothed trend line. Requests were low and sporadic through 2018, rose modestly 2019–2022, then climbed sharply after ChatGPT's release in late 2022, reaching a trend of ~10/month by 2026.

060

John Paul Helveston @jhelvy.bsky.social · 27/08/2026

Super proud of Dr. @pingfanhu.bsky.social, who successfully defended his dissertation today! In addition to writing his full dissertation, he created a storybook version that is so well done - check it out! pingfan-hu.github.io/2026-phd-sto...

pingfan-hu.github.io

A Million Cars Come Home

2101

John Paul Helveston @jhelvy.bsky.social · 24/08/2026

👂Got an article you wanna listen to? Try this tool! It extracts text from a URL or document, converts it to a narrated MP3, and posts it to your own podcast feed you can subscribe to on your phone. One line and you get a new "episode" on your phone: earmark publish <DOC> jhelvy.github.io/earmark/

jhelvy.github.io

earmark

Turn any document or article into an MP3 you can listen to.

1222

John Paul Helveston @jhelvy.bsky.social · 20/08/2026

Your opportunity to comment on the proposed NSF Guidance on Financial Assistance ends next Monday, August 24. There are WAY fewer comments on this than the earlier proposed OMB government-wide regulation, but this rule codify's the SAME policies! Attached is my comment to NSF.

030

John Paul Helveston @jhelvy.bsky.social · 14/08/2026

Thanks to everyone who joined our workshop today on Agentic Workflows with Claude Code! In case you missed it, we just posted the recording on YouTube: youtu.be/kX1bzUHZWqc Happy Claude Coding!

youtu.be

Agentic Workflows with Claude Code, w/JP Helveston & Pingfan Hu (Aug. 14, 2026)

YouTube video by John Helveston

021

John Paul Helveston @jhelvy.bsky.social · 11/08/2026

🚨 We're running a free 2-hour remote workshop this Friday on agentic workflows with Claude Code 🤖 If you're curious about how to get started with agentic workflows, you'll probably find this helpful. All workshop materials (and link to register) are at pingfan-hu.github.io/2026-agentic...

pingfan-hu.github.io

2026 Summer Workshop

Agentic Workflows with Claude Code, a hands-on workshop.

050

John Paul Helveston @jhelvy.bsky.social · 25/07/2026

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

Quote from article:
"...science. But let me make something clear right up front. When I talk to people about this subject (and that includes some journalists as well) and say what I'm about to at the end of this paragraph I am sometimes met with disbelief. But I mean what I say and I am trying my best to say what I mean: The Trump administration hates academic science funding, full stop. They hate where that money goes, and they hate who it goes to. They want to keep all that money for themselves, to hand out to favored cronies who can help them get elected and to steer yet more money and more power back into their hands.
That's it. That's the story. They are naturally not putting it in those words, but instead talking about "gold standard science" and the "reproducibility crisis" and funding "bold inquiry" for "transformational breakthroughs". This is all bullshit, and it's important to realize that. Shameless hand-waving bullshit, delivered in a how-dare-you-..."

0153

John Paul Helveston @jhelvy.bsky.social · 18/07/2026

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...

jhelvy.com

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.

060

John Paul Helveston @jhelvy.bsky.social · 14/07/2026

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...

regulations.gov

Regulations.gov

060

John Paul Helveston @jhelvy.bsky.social · 06/07/2026

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...

logitr just got even faster — and this chart proves it. Across every test, from 50 to 1,500 random draws, logitr stays almost flat near the bottom of the chart while other popular choices for mixed logit estimation climb steeply, taking many times longer as complexity increases. Even with extra cores thrown at the problem, other packages can't catch up to logitr running on a single core. Simply put: less waiting, more modeling. 🚀

1180

John Paul Helveston @jhelvy.bsky.social · 30/06/2026

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.

0133

John Paul Helveston @jhelvy.bsky.social · 30/06/2026

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?...

youtu.be

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

YouTube video by Perimeter Institute for Theoretical Physics

040

John Paul Helveston @jhelvy.bsky.social · 24/06/2026

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...

pingfanhu.com

Agentic Skills – Pingfan Hu

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

0172

John Paul Helveston @jhelvy.bsky.social · 18/06/2026

Take 15 minutes: 1️⃣ Comment using the guide (it has an AI prompt to help you draft it): defendfederalresearch.org 2️⃣ Send the same message to your members of Congress ⏰ Deadline: July 13, 2026. Then repost this.

defendfederalresearch.org

Comment on the OMB Rule

The proposed OMB Regulation for Federal Financial Assistance would put political appointees over peer review and let agencies cancel active grants. Here’s what it changes and how to comment before Jul...

000

John Paul Helveston @jhelvy.bsky.social · 18/06/2026

This is the most significant threat to American science in my lifetime. But it's not final. Public comments legally must be considered and responded to — a specific, personal comment genuinely counts.

111

John Paul Helveston @jhelvy.bsky.social · 18/06/2026

The rule would also: • Restrict international collaboration • Sharply raise the barriers to pay publication fees or attend conferences • Ban entire fields of inquiry on ideological grounds • Lock it into binding regulation that's hard to undo — and limits future public input

100

John Paul Helveston @jhelvy.bsky.social · 18/06/2026

The rule would: • Put political appointees in charge of what gets funded — effectively ending scientific peer review • Let agencies cancel active multi-year grants anytime "priorities" shift, no justification required (yes, the language is that vague)

100

John Paul Helveston @jhelvy.bsky.social · 18/06/2026

🚨 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 🧵...

defendfederalresearch.org

Comment on the OMB Rule

142

John Paul Helveston @jhelvy.bsky.social · 04/06/2026

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

Washington Post article screenshot headlined "Trump appointees push $250 banknote with his portrait." The subheading notes that the printing director who resisted was reassigned, writing "The buck stopped here" in her farewell. Dated May 28, 2026. Below the article text is a photo of a mock-up $250 Federal Reserve Note featuring a portrait of Donald Trump on the front, labeled "United States of America" and "America Anniversary". A caption states it was obtained by The Washington Post and provided to Bureau of Engraving and Printing staff in August.

230

John Paul Helveston @jhelvy.bsky.social · 02/06/2026

Yes, and most gas car drivers pay only about $95/year in gas tax. The proposal is for far more than that, while drivers of hybrids pay far less tax, yet there's no penalty for them.

000

John Paul Helveston @jhelvy.bsky.social · 02/06/2026

You would generate a similar order of magnitude in revenue by increasing the gas tax by $0.01/gal, which comes out to about $5 more per gas vehicle annually. A $130 charge per EV per year is about 6x more for the EV driver than that.

010

John Paul Helveston @jhelvy.bsky.social · 02/06/2026

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 🤔

A line chart titled "BUILD America 250 proposed EV fee penalizes EV owners & is not justified by gas tax revenue loss." The x-axis shows Gas Vehicle Fuel Economy in MPG (10–55), and the y-axis shows Breakeven Annual Miles (0–35,000). Two upward-sloping lines show the annual miles an EV owner would need to drive to pay the same in gas tax as a comparable gas vehicle: one for the $130 initial fee and one for the $150 fee cap. A horizontal dashed line marks the average annual miles driven (11,484), and a vertical dashed line marks the average MPG (22.3). The region above the breakeven lines is shaded red and labeled "EV Fee > Gas Tax"; below is shaded green and labeled "Gas Tax > EV Fee." At average MPG and average miles, both breakeven lines fall well above the average driving line, indicating most EV owners would pay more under the proposed fee than a typical gas driver pays in gas tax. Source: Federal gas tax 18.4¢/gal; BUILD America 250 Act, H.R. 8870, §1129.

5318

John Paul Helveston @jhelvy.bsky.social · 27/05/2026

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...

jhelvy.com

The Unreasonable Effectiveness of Quarto – jhelvy.com

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

1142

John Paul Helveston @jhelvy.bsky.social · 14/05/2026

Quarto is 4x more efficient on tokens than HTML outputs from Claude, but can render to the exact same HTML! And 7x more efficient than Word docs!

050

John Paul Helveston @jhelvy.bsky.social · 13/05/2026

#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...

jhelvy.com

The Unreasonable Effectiveness of Quarto – jhelvy.com

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

2324

John Paul Helveston @jhelvy.bsky.social · 03/05/2026

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

180

John Paul Helveston @jhelvy.bsky.social · 30/04/2026

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.

082

John Paul Helveston @jhelvy.bsky.social · 10/04/2026

@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...

pingfanhu.com

Agentic Engineering - Part 1 – Pingfan Hu

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

2122

John Paul Helveston @jhelvy.bsky.social · 09/04/2026

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

A map showing Vietnam (outlined in red) overlaid on Iran (outlined in blue) for size comparison. Vietnam fits well within Iran's borders, illustrating that Iran is significantly larger in land area than Vietnam.

060

John Paul Helveston @jhelvy.bsky.social · 29/03/2026

While this specific case is misleading, there's no doubt the content of the article points to a very real threat, one that we may not even be able to detect. That is, LLM responses may already be riddled throughout our data (possibly even in this case) and we may not even be able to tell.

020

John Paul Helveston @jhelvy.bsky.social · 28/03/2026

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...

theguardian.com

‘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

061

John Paul Helveston @jhelvy.bsky.social · 22/03/2026

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

030

John Paul Helveston @jhelvy.bsky.social · 11/03/2026

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

1132