Hi {{first_name|friend}},
AI has gotten a lot better at facts this year, and one of the sharpest people writing about it says we can ease up on checking its work. I agree with him, up to a point.

My Take
Ethan Mollick is right that AI has gotten a lot tighter about pure facts.
Mollick, a Wharton professor who studies AI, said this week that there's "way too much concern" about checking AI's work for factual errors. That used to be a real risk, he says, but newer models now beat human experts, including PhD specialists, at finding facts and checking claims.
I still advise folks to evaluate the rest of the response (sound arguments, voice, deep context, industry fit), especially for strategic work.
A Google researcher showed this week what that looks like when it goes wrong. He set 48 AI agents loose on NASA mapping data, and they flagged what looked like an unrecorded Maya village in Mexico. The site was real. It had also already been published in an academic journal last year, which an archaeologist would likely have checked first.
Here's what the four checks look like on your own work. Say you ask AI to recommend a price for a new service. Every number in the answer can be right and it can still miss. Does the reasoning lead to that price, or does it just sound sure of itself? If it's going to a client, does it sound like you? Does it know this client has pushed back on price before, or that your best customers come from referrals? Usually it doesn't, because you never told it. And does the margin fit your industry? What's normal in software can be wishful thinking for a contractor.
AI can confirm a date or a statistic in seconds, but whether an answer fits a client you've known for years is something you're better placed to judge. That's why I'd keep reading every strategic answer with all four in mind, however good the models get at facts.
Brett
Try This Today
Run the four checks on one AI answer you've already used. It takes about 10 minutes.
Pick the last AI answer you used for something that mattered, like a client email or a plan.
Read it once just for the argument. Does the reasoning lead to the conclusion?
Read it again for voice and context. Does it sound like you, and does it know what you know about the situation?
Ask whether it fits how things work in your industry. Fix what fails, and add the missing context the next time you ask.
Watch: why your manager may trust you less when you use AI
Worth Knowing
Doctors want protection from blame when AI makes a mistake
The American Medical Association is calling for federal and state laws that protect doctors when an AI tool, used as intended, harms a patient. It wants AI makers to carry liability insurance and disclose their tools' limits, and it says hospitals that require a tool should share responsibility. One worry it raises: an AI can make a plausible recommendation based on incomplete lab results that the doctor can't see.
If AI is part of your care, it's fair to ask your doctor whether a recommendation came from a tool and what it was based on. Read more
Nvidia's CEO resigned, and it may have touched your retirement account
Jensen Huang resigned as CEO of Nvidia after reports of two affairs, including one with a woman US intelligence identified as a Chinese intelligence asset. CFO Colette Kress is interim CEO. Nvidia has lost about $1.4 trillion in market value in two weeks, and it's one of the largest holdings in major index funds.
If you own an index fund, you likely own some Nvidia, so this may explain why your balance moved this week. Read more
A tech founder argues feeling is the one thing AI may never learn
Regan Peng, an AI founder, opens with an old voicemail from someone you've lost: you know every word and you listen anyway. AI might explain that attachment and even predict your response, he writes, but that isn't the same as knowing what it is to miss someone. He wants AI built to give people more room to live, like an assistant that notices loneliness and helps strengthen your relationships instead of keeping you hooked on the app.
I particularly like what he said about humans: "We do not have to remain the most intelligent beings imaginable for our lives to matter." Read more
Tell me
What do you still check before you trust an AI answer? I love to read your replies.
30% off all my courses, Nov 16 to 22 (my lowest price ever)
Maven's Fast Track sale takes 30% off everything I teach, Monday Nov 16 through Sunday Nov 22. The discount is applied automatically at checkout.
Level 1: Set Up AI to Work for You, live. Pick Dec 2 to 18 or Jan 13 to 29, with 1.5 hours one on one with me.
$695, $486.50 during the saleLevel 1, self paced. The same course on your own schedule, start anytime.
$495, $346.50 during the saleSkills workshop, Tue Dec 8. Build skills for your recurring work, live.
$200, $140 during the sale
The sale ends Sunday, Nov 22 at 8pm Pacific / 11pm Eastern.
