Concerning the “AI will kill us all” headlines breaking into the mainstream media: BE NOT AFRAID. The best way out is straight through; let’s not wallow in the messy middle of the Singularity. Apologies to Robert Frost. www.semafor.com/article/09/1...

View: We have nothing to fear but AI fear itself

Blindly believing advocates who say AI could kill us means missing some real risks, like economic or societal destabilization as AI transforms how people work and communicate.

semafor.com

This is what "vibe genealogy" looks like. A family historian fed his tree to Claude Code and got what genealogy software never gave him: a map of 8 generations of ancestral birthplaces, every parent-to-child move drawn as a line. Nearly all born within 40 miles of Mount Airy, NC. 🧵

I am not interested in making Claude Fable 5 and GPT-5.6 fight for a trophy. I want one to analyze the genealogical record and the other to find the holes—using the same free GRA v9.2 methodology—before I return to the actual evidence and decide. Desktop setup + workflow guide in the reply.

Free Fable 5 access ends TONIGHT, 11:59:59 PM Pacific. Last night, on vacation, I watched two rival AIs grade each other's genealogy homework: 155 records, a sealed answer key, bars set before I peeked. Zero invented facts. Four real bugs caught. One honest failing grade.

Claude Fable 5 is back — but only included through July 7, then it goes metered. Don't waste it on chat. I published a working guide: how to actually spend your Fable week on genealogy — copy-ready prompts, a live demo on my own family records, and a try-it-yourself exercise. 🧵

Claude Fable 5 is coming back — Anthropic says the export controls are lifted and access returns tomorrow. I had it for three days in June and put it to real genealogical work. Just published a field report on what a frontier model can build for family history. 🧵

Anatomy of a Shakedown. Trump floated taking "pieces" of AI companies. Days later, a national-security order forced Anthropic to switch off its two most powerful models worldwide—even its own foreign-national staff—on 90 minutes' notice. Security, or leverage?

Trump did not merely endorse AI regulation. He publicly discussed taking “pieces” from AI companies so the public could become “very rich,” framed it as a “partnership with the American public,” then said he expected top AI executives to agree to “giving back.”

1/ Trump muses that Washington should take ownership stakes in AI firms. Days later, a national-security export order forces Anthropic to switch off its two most powerful models. Stated reason: security. Real result: leverage. A chokehold.

The era of the "free" AI lunch is ending. For 3 years, VC covered the gap between your $20 subscription and what frontier models cost. Exhibit A: Claude Fable 5 goes pay-per-token after June 22 — $10/M in, $50/M out. The new skill isn't using AI. It's knowing when the expensive model is worth it.

AI is not neutral. That's the central claim in Pope Leo XIV's new encyclical: every technology carries the values of those who build it, fund it, regulate it, and use it. One of the most consequential AI statements we'll see this year — and it didn't come from Silicon Valley.

I switched my main AI app from Antigravity to Codex six weeks ago. Today I'm ~60% Codex, 30% Claude Code, 10% Antigravity—and loyal to none of them. Don't fall in love with your chatbot. Then I had Claude explain my own words back to me three ways. It got weird. 🧵

Met Kurt Luther (Virginia Tech, creator of Civil War Photo Sleuth) and his PhD student last week — came away impressed. Their lab is recruiting genealogy hobbyists for a 2-week remote study of Genet, their AI browser extension. (1/2)

Ep 41 is live: "More Is Not the Same as Better" 400x more DNA data. 2.3M searchable pension pages. Wikipedia bans AI content 44-2. And an interview with FamilyTreeDNA's David Vance that's one of the best we've done this year. podcast.show/3738800/epis...

In 2023, AI could handle a 4-minute task. By 2026, 12 hours. That's the METR time-horizon benchmark — and the curve is nearly vertical. One metric, not the whole story, but hard to look away from. And new models coming soon.

Fair warning: "I built a thing" post incoming. 18 months. A free prompt that teaches AI the difference between "primary source" (wrong) and "original source containing primary information" (genealogy).

Friend asked if anyone had a tool to turn any file into a GEDCOM file. Built one in a couple of hours. Tested it on a 1909 newspaper article -- both Claude and ChatGPT caught a contradictory marriage date I hadn't noticed. Free prompt, works in any chatbot or as a skill in agent harnesses!

Download this Virginia seal “No Kings” image in multiple sizes and aspect ratios—square, portrait, landscape, and Facebook banner—and read the full story behind it and the March 28, 2026, rally at (link in first comment):

Virginia put the answer on its state seal in 1776: Virtus standing over a fallen tyrant, his crown in the dust, his chain broken. Sic Semper Tyrannis. Thus always to tyrants. This isn't politics. This is inheritance. March 28. #NoKings

Fun Prompt Friday: I ran a 108-line image analysis prompt on the same family photo across four AI models. Claude, GPT-5.4, Gemini, Grok — scored head to head. Free prompt, full comparison matrix, and the results are not what I expected. #genealogy

Ten days. The president's counterterrorism director quit, saying this war was started under foreign pressure. Courts ordered Voice of America restored. Thirteen Americans dead, no congressional vote. Now is the time for all patriots to come to the aid of their country.

Thirteen days. Now is the time for all patriots to come to the aid of their country.

I made a quiz that maps you onto a spectrum of 34 real AI thinkers. Takes two minutes. You get a personalized report — allies, blind spots, tensions, reading list. Then I wrote about why genealogists should be building their own tools, not buying them. And five examples to try today. Link below.

Just published my RootsTech 2026 takeaways. Five things that stood out — including why Alton Brown's "uni-tasker" rule applies to AI genealogy apps. Also: I'm more direct than usual about our Academy course. Fair warning. Link in reply.

Fresh off RootsTech 2026 — Mark Thompson and I taught six sessions on AI and genealogy. Now we're teaching a 5-week Level I course starting March 17. Practical AI skills + a community that lasts. Recordings available 60 days. 15% off: ROOTSTECH15OFF Link in reply.

Anthropic CEO draws two red lines with the Pentagon: no mass domestic surveillance, no fully autonomous weapons. The Dept. of War threatens retaliation—supply chain blacklist, Defense Production Act. Anthropic won't budge. This may define how military AI is governed for years.

Claude Skills are the next iteration or evolution of prompt-based genealogy work. Nicole Dyer is a Jedi Master of explaining this technology in a genealogical context.

A karaoke company erased $17.4 billion from the stock market. Look at this chart — AI capability just went vertical. "Hard take-off" is the scariest term in AI right now, and your feeds are full of it. Here's what it actually means and who's saying it.

Is AI boiling the oceans? The popular narrative is off by orders of magnitude. As Ethan Mollick warns, an "omnicausal anti-AI" coalition is forming — lumping every fear together and crowding out the nuanced policies we actually need. 🧵👇

This is the strangest story you'll read all month. It's not the birth of Skynet. But it may be the beginning of Facebook for Baby Skynet. Since the AI Revolution kicked into third gear late last year, the pace of acceleration has been picking up.