A week that broke the internet

Two AI stories collided: a fight over whether OpenAI quietly borrowed others' work on a million-dollar math problem, and an Anthropic researcher quitting over safety.

Ten percent, with an exclamation point

An Anthropic staffer said colleagues earnestly believe AI could kill all humans, putting it above 10 percent within the decade. The debate over whether that is real or theater went everywhere.

“I personally think it is >10% within the next decade.”

Who Gebru is

Timnit Gebru was hired by Google in 2018 to study bias in its AI. Google rejected her paper on the dangers of large language models, and she was gone soon after.

Why math, of all things

Gebru asks why labs pour money into math, chess and programming. Solve those and you get to announce you have solved intelligence.

“It's so they can say, 'We solved it.'”

The vetting never happened

Math has a process for checking novelty and credit. Waiting for it would have shown whether OpenAI described its breakthrough accurately.

Press release to policy

The Leiden Declaration warns what happens when corporations use mathematics this way. Policymakers read press coverage instead of asking mathematicians, and bills follow fast.

“The amount of time it takes now to go from companies claiming an AI breakthrough to Bernie Sanders proposing a bill is really short.”

IPOs killed collaboration

At Google, Gebru's team worked openly with researchers at Microsoft and Amazon. Now everyone is frantic and closed, racing toward public offerings.

Can a bridge decide to collapse

Gebru's line from her upcoming book. Nobody asks whether a fallen bridge was sentient or ethical. You ask who built it flimsy, and who skipped the permits and tests.

“When you ask about the bridge collapsing itself, you've lost the thread.”

The harms already here

Autonomous weapons in active warfare. Climate damage from the buildout. Bosses using AI as the excuse to cut staff. All real, all now.

“This machine-god narrative is, in my opinion, meant to distract us from these.”

A privilege of the safe

Extinction scenarios stay vague: maybe it hypnotizes you, maybe it spreads malware. Gebru notes we already have malware, and that obsessing over the machine-god takes a life untouched by police violence, floods or war.

“Does an angry GPU just show up at my house or something?”

Same script since the fifties

Gebru reads people old AI quotes and asks them to guess the decade. They often cannot, because it all sounds the same.

“The singularity has been coming for decades now.”