谷歌告诉几百万人：Sam Altman 死了。消息来自维基百科上一次只存在了四十一分钟的篡改。

On the twelfth of August, people searching for Sam Altman on Google saw his death notice as the first line.

Sam Altman was not dead — nothing had happened to him.

That day, he was at OpenAI as usual.

It began with an act of vandalism.

Wikipedia has an entry for him, and Wikipedia is editable by anyone.

That day someone opened his page and added a date of death, a city of death, and a more serious claim about the cause.

Not one line of it had any basis.

Forty-one minutes later, another editor spotted the vandalism and deleted all of it.

Forty-one minutes, set against the fourteen hundred-odd minutes in a day, is a very short window.

But within those forty-one minutes, Google’s Knowledge Panel scraped it.

Google did not look for a second source to corroborate it, nor did it set a higher bar for a claim that somebody had died.

It displayed that death notice directly at the top of the search results.

So that day, many people searched the name and the first thing they saw was that he had died.

Once the story spread and the backlash came, Google took it down.

Google’s response afterwards is the part actually worth chewing on.

Google said: this was not manually changed by our people.

That statement is accurate — no Google employee did touch it.

But read it in reverse and it means: nobody wrote it, and nobody looked at it.

A stranger typed a few lines on a public page, an automated system took it as fact, and forwarded it to millions.

Whether or not a person has died leaves traces everywhere else.

That system saw a single source, and that was enough.

The problem is not that the model turned bad.

The problem is how easily it believes.

Think about it: of the things you believe today, how many has somebody really verified for you?