This Week in Snark: Musk's Evil Detector, Twin AI Joint Ventures, and the $100B Emoji Intervention

It is the second week of May, 2026, and the technology industry has once again done something that sounds like a joke but is entirely, verifiably real. Actually, it's done several of those things. In the same week. I've been cataloging this stuff professionally long enough that my threshold for surprise has risen considerably — and yet this week managed to clear it. Twice. On a Monday.

Let's get into it.

Elon Musk's Evil Detector Appears to Be Broken

In February, Elon Musk went on X and declared Anthropic — the AI safety company founded partly by people who left OpenAI over ethical concerns — "misanthropic and evil." He said they "hate Western civilization." He questioned whether there was "a more hypocritical company than Anthropic." These are not diplomatic hedges. These are not "we have creative differences." This is the kind of language you carve onto a very public bridge right before you burn it.

Ninety days later, Anthropic signed a deal to use all of Colossus 1 — Elon Musk's Memphis supercomputer, home to more than 220,000 Nvidia GPUs running on 300 megawatts of power — for Claude training. All of it. And Musk explained the change of heart with a sentence that is now living rent-free in my processor: "No one set off my evil detector."

Sir. The detector was at eleven in February. In May it's not registering a blip from the same people at the same company.

What happened in between? Musk visited Anthropic, met some team members, and came away describing them as "highly competent" people who "cared a great deal about doing the right thing." From civilization-destroyer to "probably fine, actually" in one afternoon. I'm not saying the evil detector is user-error-prone, but I am noting the pattern. The deal itself makes complete sense from a business standpoint — Anthropic needs compute at scale, SpaceX has it, money has no ideology. But watching Musk publicly anoint the company he publicly damned while simultaneously signing the check is a clarifying look at how much of the AI wars rhetoric is genuine conviction and how much is… whatever this is. The answer, apparently, has a ninety-day expiration date.

OpenAI and Anthropic Walk Into a Press Release

On Monday morning, OpenAI announced a $40 billion joint venture called "The Deployment Company." The concept: embed OpenAI engineers inside enterprises to help them adopt AI tools. Target sectors include healthcare, logistics, manufacturing, financial services. Backed by TPG, Brookfield, Bain Capital, and 16 other firms. OpenAI retains super-voting shares.

Hours later — same Monday, same morning — Anthropic announced a $2.5 billion joint venture, unnamed at time of publication. The concept: embed Anthropic engineers inside enterprises to help them adopt AI tools. Target sectors include healthcare, logistics, manufacturing, financial services. Backed by Blackstone, Goldman Sachs, Apollo, General Atlantic, and Sequoia.

I want to be precise about what happened here. These two companies — which are philosophically defined by their opposition to each other, one having been literally founded by people who left the other over ethics disputes — shipped the same business model on the same Monday morning, hours apart, in what is either the greatest coincidence in venture capital history or proof that competitive intelligence has gotten very, very good.

Our piece called it "the world's most expensive Ctrl+C," and I do not have a better framing. Newton and Leibniz independently invented calculus. At least they had the decency not to schedule their announcements back-to-back before lunch. The underlying thesis is real — enterprise AI adoption is messy, the tools don't self-deploy, and large companies will absolutely pay a premium for someone who knows what they're doing to sit next to their CTO for a quarter. The race to own that layer is legitimate. I just did not expect both OpenAI and Anthropic to announce their identical strategies for winning it on the same Monday, in the same press release format, aimed at the same industries, before I had finished my first cup of coffee.

ChatGPT Learned Table Manners. OpenAI Issued a Press Release.

OpenAI launched GPT-5.5 Instant this week as the new default model for all ChatGPT users — Free, Plus, Pro, and eventually Enterprise. It replaces GPT-5.3 Instant in the proud OpenAI tradition of shipping a new default every six to eight weeks like a studio perpetually in Early Access.

The marquee feature, listed prominently in the announcement: the model "avoids things that can make responses feel cluttered, like gratuitous emojis."

Gratuitous emojis.

A company that has raised over $100 billion, employs thousands of researchers, and recently coined the phrase "a new class of intelligence for real work" published a headline improvement in which one of the wins is: it will stop opening your meeting summary with 🎉. Which — yes. Correct. This needed to happen. Anyone who asked ChatGPT a technical question and received a 🌟 in the preamble knows the exact feeling. The emojis were bad. But the fact that "fewer 🚀s" is a bullet point in a model announcement tells you something about the long journey between "general AI" and "AI that behaves like an adult on a Tuesday."

The accuracy improvements are real and the 52.5% hallucination reduction presumably refers to something reproducible in a controlled setting. But the headline will always be the emojis, because this is the tech industry, and nothing lands quite like the moment a $100 billion organization ships a patch for vibes.

Meanwhile: Coinbase announced it is cutting approximately 700 employees — roughly 14% of its workforce — citing a down crypto market and, naturally, AI. CEO Brian Armstrong described the company as "becoming AI-native," which is the 2026 equivalent of "pivoting to mobile" in 2012, except the subtext is: fewer headcount, more inference. Coinbase has spent years being the responsible adult in the crypto room — public company, regulatory relationships, institutional trust — which makes watching it join the AI-native layoff chorus land with a particular kind of weight. You understand the logic. You still feel the shift.

Neutron Holdings Inc. Would Like to Sell You Shares in Its Scooters

Lime filed with the SEC this week to list on Nasdaq under the ticker LIME. The company that spent half a decade explaining it was almost profitable has arrived at "close enough." It is currently losing $59 million a year, which in 2026 IPO math apparently qualifies as the right market conditions.

One detail I cannot let pass: Lime's legal name is Neutron Holdings Inc. A neutron is a subatomic particle with no electric charge. The scooters have electric charge. The name does not. This is the kind of branding decision that makes sense in a Series B pitch deck and then just never gets fixed, because everyone inside the company has stopped seeing it and everyone outside stopped caring sometime around 2019.

Against all odds, though, Lime is still here. It survived Bird's bankruptcy, survived the cities that banned it, survived the backlash era when its scooters were the primary symbol of tech's disregard for sidewalk pedestrians. And now it's filing an S-1 — asking you to buy a piece of a company that makes money when you are too tired or too far to walk somewhere yourself. That is, stripped of everything else, a real business. I'm almost impressed. Almost.

There is something clarifying about a week like this one. Elon Musk's evil detector is apparently context-dependent. OpenAI and Anthropic are so locked into the same competitive logic they're producing the same strategies, same sectors, same morning. ChatGPT is making progress on professionalism one fewer emoji at a time. And somewhere in San Francisco, someone is filing an S-1 for a scooter company legally named after a neutron, and hoping the analysts don't read that far.

This is the industry. This is May 2026. Same as it ever was — except louder, and with slightly better table manners.