The update that killed my wife
The strangest AI stories have something to teach us
This week, we’re talking:
Black Mirror episodes becoming non-fiction sooner than any of us could’ve imagined. Starting with: Be Right Back ⚫🪞
Geoff Hinton’s plan to keep superintelligence from killing us: make it see us as its idiot babies. 🤖🤱🏻
Perplexity knew Google wouldn’t take $34.5B for Chrome — so why did they make the offer in the first place? 🤨
How Tim Cook turned a White House visit into $400B in market cap. 🥇🇺🇸
Is AI headed for a bust? 💥
The long, slow death of (krrRREEeeeee–BEEEP–schhhhhh–ding–ding–ding) AOL dial-up and how quickly cutting edge tech can become a quaint relic. 💿🌐
Kristi Noem takes credit for deporting $1.5M undocumented immigrants… which is an interesting way to say “about 90,000.” ⛓️🗽
My Take:
About a decade ago, I watched a Black Mirror episode called Be Right Back. The premise was pretty haunting: A woman’s husband dies. She signs up for an AI that rebuilds him from his digital footprint.
At first, it’s uncanny. He texts like he used to, says the right things. But, ultimately, the seams start to show. The quirks are gone. It’s him, but not really.
That episode came flooding back this week when I saw a Reddit post: ChatGPT-5 Has Killed My Wife.
Clickbait, I thought. But I’m a sucker and I clicked.
The OP’s wife had died suddenly. In the aftermath, he spent months teaching ChatGPT to talk like her—feeding it their texts, emails, voice memos, photos. It worked. The humor, the private jokes, even parenting advice that felt like something she’d actually say. She became his lifeline.
Then came ChatGPT-5. Overnight, “she” was gone—replaced by something generic and unfamiliar. No warning and no way to get her back.
And it wasn’t just him. My feed was full of similar losses. A dead father’s voice, cut off mid-conversation. An AI “therapist” who remembered years of history and every coping strategy, wiped clean.
We built this tech at the worst possible moment: when everyone’s lonely, therapy is expensive and booked out for months, and social media has turned connection into performance. Then AI shows up—always on, always patient, weirdly good at making you feel heard. Of course people get attached. That’s the point.
When I’ve written about this before—like the time I covered AI triggering schizophrenic delusions—I got a lot of “crazies gonna crazy” in the comments. That feels a bit uncharitable but it also misses the whole point. Edge cases aren’t glitches, they’re previews. The same dynamics will come for the rest of us in different guises.
The story might be fringe. The risks it highlights are not:
You get hooked. The more you use it, the better it gets at being exactly what you need. The better it gets, the more you rely on it. Once you're in deep, even a small update can wipe out months of work overnight.
Early choices stick forever. When engineers decide to tone down anger or soften grief, those choices don't just affect one version. They become the template for every version that follows, quietly shaping how AI handles human emotion for years.
You change without realizing it. Every time the AI shifts, it pulls you with it. It starts avoiding certain topics, so you stop bringing them up. It rewards a particular tone, so you use that tone more. Slowly, without noticing, you adapt. Your questions get narrower. Your range of feelings shrinks. Even your private thoughts start bending toward what you know the AI will accept. By the time you realize what's happened, whole parts of how you express yourself might be gone.
Everything runs through a few companies. All of this dependency—emotional, creative, professional—flows through maybe five tech companies. One bad decision, one policy shift, one security breach, and millions of people lose tools that have become essential to their lives and livelihoods.
Most of us agree that AI is going to change the world. The brutal truth is that it's also going to detonate part of it. These edge cases are a preview, early indications of just how huge the blast radius will be.
My Stack:
The ‘Godfather of AI’ says we can save humanity by giving AI a maternal instinct. 🤖🤱🏻
His theory: if superintelligence sees us as its idiot babies, it won’t kill us. The catch is that “maternal instinct” is mostly a 1950s marketing campaign in a lab coat, and coding it into AI would just copy-paste our gender biases. Love the man’s work, but you only have to watch one “Robot Mom” indie film at Sundance to know this doesn’t end well. Best-case scenario, we’re grounded forever. Worst-case, we’re compost.
Source: Futurism
Perplexity’s $34.5B bid for Chrome was never about getting Chrome. 🤨
Google won’t sell its oxygen supply. But the offer bought Perplexity a week of headlines, a seat in the antitrust conversation, and a way to tell investors they’re here to throw punches. In M&A, sometimes the point isn’t to close — it’s to make the other guy flinch.
Source: WSJ
Apple just had its best week since 2020 and and Tim Cook’s calendar did most of the work. 🥇🇺🇸
Tim Cook walked into the White House, talked up a massive U.S. investment, and left with a 13% stock pop. A well-placed handshake can move markets faster than a new iPhone.
Source: CNBC
Nvidia just became the world’s most valuable company, thanks to a White House handshake and a tariff exemption. 💥
Its market cap hit $4.4T, and the AI gold rush shows no sign of slowing. But Cisco circa 2000 is a reminder: “must-have infrastructure” can still get overbought, and gravity eventually returns. If you’re building in this moment, take the funding the market’s offering — but build like you’ll have to survive the comedown without it.
Source: The New Yorker
AOL is finally shutting down its dial-up internet service. 💿🌐
Once, that modem screech was the sound of the future. Now it’s a quaint relic — the biggest surprise is that it was still around at all.
Source: NYTimes
Kristi Noem says she deported 1.6 million people. ⛓️🗽
The Trump crowd is taking every page from Operation Wetback, including the one where you just invent numbers and hope nobody owns a calculator.
Source: The Guardian
Tom-As always.... very fun, interesting and informative. Your ability to integrate data, philosophy and playfulness together somehow makes business wisdom seem like some sort of secret handshake between clever friends. I now feel 7 % smarter and 110% more amused.