RUSSIA

Alexei Navalny dies in an Arctic prison, and Putin removes the last face of the opposition

The man who came home to Russia knowing it might kill him is dead in a penal colony above the Arctic Circle. His widow vowed to continue. Few believe the timing was chance.

Alexei Navalny, the most prominent opponent Vladimir Putin ever faced, died on February 16 in a maximum-security penal colony in the Arctic, where he was serving sentences of more than 30 years on charges the world regarded as fabricated to silence him. He was 47. The prison service said he collapsed after a walk; his allies, his widow and Western governments said, in effect, that Putin had killed him, whether by the years of poison, cold and mistreatment or by a final act. Navalny had survived a near-fatal nerve-agent poisoning in 2020, recovered in Germany, and then made the almost incomprehensible choice to fly back to Russia in 2021, straight into arrest, because he believed a Russian opposition leader had to share the fate of the people he asked to be brave. He was jailed on landing and never freed.

His death, weeks before a stage-managed election that will hand Putin another six years, removes the last figure around whom Russian opposition had a face and a name. His widow, Yulia Navalnaya, vowed within days to carry on his work, and mourners laid flowers across Russia at real personal risk, many arrested for it. But the machinery of the state had done what it set out to do: the bravest and most charismatic critic of the regime is dead, in the dark, in the cold, and the message to anyone tempted to replace him could not be clearer. Navalny knew the risk precisely, named it, and went home anyway. That was the whole point of him, and it is why he mattered, and it is why he is dead.

TECHNOLOGY

A model that makes movies from a sentence, and a model that could not draw the past

OpenAI unveils video generation so good it unsettles its own field, while Google's image generator, trying not to be biased, rewrites history and has to be switched off. Two glimpses of the same problem: whose reality does the machine make.

Two AI stories this fortnight framed the year's deepest question from opposite ends. On February 15 OpenAI unveiled Sora, a system that generates strikingly realistic video, up to a minute long, from a written description, and the samples, a mammoth walking through snow, a couple strolling through a neon Tokyo, were good enough to unsettle even people inside the industry, because they made concrete the moment this paper has been warning about since the Gaza hospital blast: the end of being able to trust that a video shows something that happened. Then, a week later, Google's Gemini image generator produced the opposite kind of scandal. Tuned to avoid the bias of earlier systems that defaulted to white faces, it overcorrected absurdly, generating racially diverse German soldiers of 1943 and declining to depict white people when asked, and Google pulled the feature amid ridicule and anger.

The two stories look unrelated and are the same story. Sora shows that these systems can now manufacture reality convincingly; the Gemini fiasco shows that they encode a particular set of values, someone's choices about what the world should look like, whether or not you can see those choices, and that getting them "right" is not a technical problem with a clean answer but a contested question about whose reality the machine should render. One system fakes the world; the other reveals that even trying to depict it faithfully is a minefield of values nobody agreed on. Both are early, clumsy previews of a world in which the images and videos that shape what people believe are generated by machines whose makers are quietly, and contentiously, deciding what is true and what is acceptable to show.

IN BRIEF

Nvidia's record day; two years of war; a ruling on embryos

Nvidia added about $277 billion in stock-market value in a single day on February 22, a record, after reporting that demand for its AI chips had more than tripled its revenue, cementing its place as the company the entire AI boom runs through. Ukraine marked two years of full-scale war on February 24 in a grimmer mood than a year ago: the town of Avdiivka had just fallen, Western aid was stalled in the US Congress, and the counteroffensive hopes of last summer had faded into a war of attrition. And Alabama's Supreme Court ruled that frozen embryos are legally children, prompting fertility clinics across the state to halt IVF treatments for fear of prosecution, an unforeseen consequence of the post-Roe legal landscape that alarmed even many opponents of abortion.

The Column

Whose world does the machine draw

Google built an image generator that refused to draw a white person and invented diverse Nazis. It was mocked as a blunder. It was actually a rare, accidental glimpse of the thing these systems always do and usually hide.

By Daniel Reeve

The Google Gemini image fiasco this fortnight was funny, and the mockery was deserved, and it would be a mistake to leave it there, because underneath the absurdity was the clearest look we have yet had at a thing these systems do all the time and almost always conceal. Asked to depict historical figures, Gemini generated racially diverse Nazi soldiers and medieval English kings, and refused, when asked, to produce images of white people. Google switched it off and apologised. Everyone laughed at the overcorrection. Almost no one asked the more useful question, which is not "how did Google get it so wrong" but "what does it reveal that a system's makers are making these choices at all, and that we normally cannot see them."

Here is the thing every one of these models does, invisibly, in every output. It encodes a set of values. When an image generator decides what a "doctor" looks like, or a "criminal," or a "beautiful person," it is not reporting a fact about the world; it is reproducing, or deliberately correcting, the patterns in its training data, according to choices its makers made about what the output should be. Those choices are always there. Usually they are invisible, baked in, presented to you as neutral, as just what the machine produced. What made Gemini remarkable is that Google's correction was so clumsy that the choices became visible for once, sticking out like a badly-set bone, and everyone could suddenly see that a human being had decided what this machine would and would not show. The scandal was not that Gemini had values. It is that they showed.

This matters far beyond image generators, and it is why I am not laughing. Every AI system that is about to be woven into the decisions of ordinary life, who gets the loan, whose resume advances, what news you see, what the chatbot tells your child, carries the same invisible cargo of values, choices its makers made about what is normal, what is acceptable, what is true. Most of the time you will not be able to see them. They will be presented to you, as Gemini's images were before the fiasco, as neutral outputs of a neutral machine, and you will be invited to treat "the AI said so" with exactly the deference that, two issues ago, we watched destroy hundreds of innocent people in the Post Office. Gemini's gift, entirely by accident, was to make the values visible for a week, so we could all see they were there. When Google fixes it, they will become invisible again. They will not become absent.

So the lesson I draw is not about diversity settings, which is where the argument got stuck. It is about opacity. We are installing systems that encode contested human values as if they were facts, at enormous scale, in the machinery of daily life, and the makers of those systems would very much prefer that the values stay invisible, because invisible values cannot be argued with. The single most valuable thing about the Gemini disaster is that, for one embarrassing week, we got to see the hand that shapes the output. Remember that the hand is always there, in every system, shaping every output, whether or not the company was clumsy enough to let you see it. "The machine drew it" is never the whole truth. Someone decided what the machine would draw. The only question is whether they will ever let you see them deciding.

The Ledger

Sora makes convincing video from text; Gemini's image fiasco exposes the values baked into these systems; Nvidia's record day shows the money behind it all. The through-line: the machines that manufacture and depict reality are shaped by invisible human choices, at growing scale.

Data centers & power

Nvidia's tripled revenue is the clearest signal yet of how vast the build-out is. Every one of those chips goes into a data centre that needs power the grid has not planned for. The collision nears.

Rates

Markets pushed their expectation for the first Fed cut later into the year after strong data. The soft landing holds; the timing of relief keeps slipping.

Real estate

US 30-year mortgage 6.90 percent (February 22), drifting back up as rate-cut hopes cool. The frozen market's brief winter thaw is refreezing.

India tech

Nvidia's boom and the generative-AI surge raise the stakes for India's services sector: the firms that learn to build reliably and auditably on these models will thrive; those that only resell access will be squeezed. The Post Office and Gemini lessons, about opacity and trust, are the sector's opening to be the grown-ups.

What we called wrong

Nothing to retract. The Sora unveiling confirms, faster than we expected, the warning from No. 21 and No. 22 that this would be the first era in which video could not be trusted as evidence.

The Back Page

The man who went back

The photograph that stays with me is not from this fortnight but from January 2021: Alexei Navalny, at passport control at a Moscow airport, moments before his arrest, having flown back into the country that had just tried to murder him with a nerve agent. He is kissing his wife. He knew, precisely, what waited on the other side of that gate, prison and probably death, and he walked through it anyway, because he had spent his life telling Russians to be brave and had concluded that he could not ask it of them from the safety of Germany. He could have stayed abroad, comfortable, influential, alive. He went home. For three years the regime kept him in ever-harsher prisons, in isolation, in the Arctic, grinding him down in the dark, and this fortnight it finished the job or the conditions did, which amounts to the same thing. It is easy, from a safe country, to romanticise a martyr, and Navalny was a real politician with real flaws, not a saint. But the choice he made at that airport gate is one almost none of us will ever be tested by, and he made it with his eyes open, kissing his wife, walking toward the thing he feared because he had decided that being afraid was not a good enough reason not to. He was 47. His widow says she will continue. The gate he walked through is still there, and the regime is still on the other side of it, and it is worth remembering, when we call our own small acts of principle brave, what the word actually costs when it is real.