The response to our AI symposium has been overwhelming—genealogists want honest conversation beyond "tips & tricks." As someone using AI daily in my genealogy work, I'm excited AND concerned. Before Wednesday's session, I've mapped the terrain: the "Atlas of AI Awfuls." Thread 🧵

For two years, AI-generated family trees were gorgeous garbage. My great-grandmother "Sessie" became "Susie." Dates shifted. Entire branches vanished. Then we figured out how to eliminate hallucinations completely. New AI. Different workflow. Perfect accuracy. #genealogy

Not everyone's #Thanksgiving table is easy. If yours requires some strategic boundary-setting, here's permission—with a side of humor. You're in good biblical company.

"Did you know? A 5-minute chat about the weather boosts synergy by 0.4%! Talk to someone today!" - Acme Corp HR #TinyPersonalUniverse

Q: Why does AI re-imagine a photo instead of editing it? A: It doesn't "restore" your photo — it paints a new one, guessing details from millions of other images. Eye color, skin tone? Fabricated. Always label. Always cite. Illustration, not evidence. #genealogy

Prompt to show DNA inheritance like mixing paint. Distant ancestors are vivid, concentrated colors—the raw pigments. As generations blend together, the colors soften. Each child is a visual mix of their parents' colors, highlighting where (from whom) the color came from, and from how long ago. 🔗➡️

Yesterday, I took a screenshot of my family tree at 5:20 PM. By 6:40 PM, I had an Nano Banana Pro-generated heirloom chart with EVERY name verifiable. Between those times? Total failure, an EXPLETIVE DELETED moment, and an insight that you can use.

You will hear of launches and rumors of launches—but here's how to test Gemini 3 yourself. Pick one stubborn document or photo. Run it through our #genealogy research-tuned Gem today (Gemini 2.5). Save everything. When G3 lands, rerun the exact same prompts. If it's real, you'll feel the difference.

Know Your Data. Know Your Model. Know Your Limits. Prof Mollick articulates why this framework matters: "You need to know specifically what YOUR AI is good at, not what AIs are good at on average." Test models on YOUR real genealogical work—not generic assumptions. Link ➡️

The 'AI slop' critics point to? That's just beginner output. Learning LLMs is like learning to ride a bike—you can't read your way to mastery. Smart people try once, produce dreck, then dismiss the tool. But they're confusing their inexperience with the tool's limits.

A Sunday morning reflection: “Every design choice expresses a vision of humanity.” —Pope Leo XIV There are better and worse ways to steer the AI revolution. The worst hands our values to a few unaccountable corporate tech boards while governments dither or sell-out. (1/4)

This spring, new AI photo tools arrived. Genealogists experimented. Then quickly we realized: these tools don't "restore" historical photos—they may fabricate new faces. By August, these "restorations" were everywhere, unlabeled, corrupting our shared historical record. #genealogy

The Coalition for Responsible AI in #Genealogy has released its new official position statement: "Protecting Trust in Historical Images." While AI "photo restoration" tools can be helpful, they also risk altering or fabricating features, misleading researchers. Link and more ►

Nearly every county in Virginia shifted Blue at the 2025 off-year election. Trump will be outraged, not chastened, and when he eventually orders forces to open fire on citizens, it will appear reasonable—even necessary—to his followers. Here's why. jstephenlittlejr.wordpress.com/2025/11/05/s...

Hello friends and fellow researchers! Episode 36 of the podcast, “A Simple Path to Better Prompts,” is out this morning, and I’d like to give you a long walk-through of the Tip of the Week, "Building Reusable Research Prompts," which I shared in this episode. (Link in next comment.) #genealogy

I've written thousands of prompts over the last three years; here are the sixteen that have stood the test of time. These are my best that I still use every day, gathered in one quick copy-and-paste guide. I keep these at my fingertips in a clipboard manager, and reuse them all day long.

Some folks think AI is a bubble. That is wishful thinking. Companies will fail, but the AI revolution will roll-on. The difference is the rate of change. Even if progress stopped today, it will take decades to discover the capabilities already under our fingertips.

Asking a model to critique or improve your prompts--called "meta-prompting"--is a strong and good use of these tools. And both OpenAI and Anthropic has tools to help you craft better prompts. console.anthropic.com/dashboard

I think this is my favorite use of ChatGPT so far. Get the guide here: www.theformidablegenealogist.com/product-page... #genealogy

Anyone can learn to use this. Get it here: www.theformidablegenealogist.com/product-page... #genealogy

The barrier between an idea and a compelling image has essentially disappeared.

Here's a prompt to add a spectral wire sculpture to an image, inspired by the context and contents of the image (link to full prompt in first comment).

Virginia’s Great Seal (1776)—devised by revolutionaries George Wythe and George Mason—shows the Roman goddess Virtus, spear grounded, sword poised, standing over a deposed, crown-fallen tyrant; its motto “Sic semper tyrannis” (“Thus always to tyrants”) proclaims unending resistance to oppression.

Ep.25 of Family History AI Show: GPT-4o’s photorealism & OpenAI’s o3 geolocation reasoning impress—accurate within blocks. Plus, Google Gemini Veo 3 hints at next-gen video/audio generation.

Some models are better with words than others. I was using OpenAI's o3 to help me understand some text, so I prompted a summary, and it responded: "ignorance is the seedbed of unintended harm"

You guys...? We did it. 🥰 We won the FULL NEW YORK STATE DEATH INDEX FOR 1880-2017! We beat the largest government agency in the state of New York (@nysdoh.bsky.social), in the highest court of the state, through a Freedom of Information Law (#FOIL) suit. www.reclaimtherecords.org/records-requ...

Ask Hard Questions! Drop your toughest AI-in-genealogy questions below. Ahead of Sunday’s NGS panel on Guiding AI Principles—Accuracy, Disclosure, Privacy, Education, Compliance—I’ll be on stage with Lynn Broderick and BCG president David Ouimette. What should we tackle? ⬇️

o3 make geolocation trivial. chatgpt.com/share/6825e9...

My improv on "Please" and "Thank you": the wink.😉

Tonight is the 250th anniversary of Paul Revere’s midnight ride. May his memory remind us all to resist the tyranny forming in our government.

#genchat More on ethics & #Genealogy

Meet Linka, an AI assistant introduced by Borland Genetics, who helps Kevin Borland code the site to help users. If you'd like to learn how beginners can create your own AI assistants, Andrew Redfern is teaching a free webinar on saved prompts. #genealogy Details on both in comments. 🔗⬇️

This is one not to be missed: learning to use, create, and share saved prompts is a valuable skill, the key to unlocking many other AI superpowers.

For I4GG AI Day, please enjoy my "AI Resources for Genealogists" infographic. Mark @makingfamilyhistory.com Thompson and I are excited to be collaborating with CeCe Moore and the Institute for Genetic Genealogy to present I4GG AI Day 2025 on Fri 28 Mar 2025, both in-person and virtually. #genealogy

Fun Prompt Friday! Prompt AI to sketch your genealogical roadmap: detailed locality guides full of the places, people, and resources shaping your family history. #genealogy #familyHistory #localityGuides Link below 🔗👇

Four major companies released AI research agents (Google, DeepSeek, Perplexity, xAI, OpenAI), but OpenAI's Deep Research stands apart in comprehensiveness, despite high costs ($20/mo: 10 queries, $200/mo: 120 queries). See for yourself! Link below 🔗👇

One of my most-used saved prompts is this copy-editing tool; it analyzes text for 10 common writing issues, suggests improvements, and puts you in control—handing you the red pen to decide which changes to keep. Link below 🔗 👇

Join APG on 2 March for a FREE webinar about the importance of the U.S. National Archives in our profession and how YOU can effectively engage with legislators to support archival access. Open to all; register at bit.ly/4hPJXRg. This presentation is underwritten by Civil War Records.

This classic advice is now two years old and still good; from Mollick's "Magic for English Majors" (2/23). #aigenealogy

Are you a journalist, genealogist or historian who has ever used records from the National Archives? It's time to step up and take some simple steps to preserve federal records. And save our nation's history. Please visit this website. And share widely. fundnara.com

Hey AI, how many people are asking, "Is Trump trying to break America?"?

"Prompting is a weird art." This characterization echoes and mirrors my experience.

Authoritarians always attack independent sources of knowledge--they want the ability to decide what is true and what is not true. Teachers, librarians, journalists, researchers are all being threatened and harassed, even Wikipedia: www.404media.co/email/38ceb1...

Be precise and exhaustive with reasoning model prompts, as if instructing an overly literal genie.

I ❤️ Archivists

This is not normal.

Celebrate Black History Month by tracing African American families through time with this FREE webinar "Beginning African American Research on Ancestry®" on Wednesday, February 5th.

OpenAI just released new AI models that "think before they speak." Unlike regular chatbots that respond instantly, these take extra time to plan their answers. They're best for complex tasks like planning and analysis, rather than quick back-and-forth conversation. Link below 🔗👇

"The release of safety reports often signals the upcoming availability of new AI models, suggesting o3's public release could be imminent."

These tools don't need to be perfect to be powerfully useful; don't let the perfect be the enemy of the good.