1
00:00:00,600 --> 00:00:09,500
The man who spent a decade warning that AI could end
humanity now says it probably will not, and that we
should enjoy the ride.

2
00:00:10,600 --> 00:00:19,500
His own estimate of the odds is up to one in five that
it ends catastrophically, a number he now files under
acceptable risk.

3
00:00:20,600 --> 00:00:29,500
His proposed safeguard is that rival labs get two
weeks with each other's models before release, and
hold a phone call every few weeks.

4
00:00:30,600 --> 00:00:39,500
That is the entire mechanism, competitors checking
competitors, with governments informed only
afterwards, and only if something looks very dangerous
indeed.

5
00:00:40,600 --> 00:00:49,500
You might expect governments to object, but in June
Washington answered with a security framework for
frontier models that is explicitly voluntary.

6
00:00:50,600 --> 00:00:59,500
The same administration has told its lawyers to
challenge state AI laws, and Brussels has pushed its
strictest rules back sixteen months.

7
00:01:00,600 --> 00:01:09,500
Because everyone is watching the same rival. China
generates more than twice America's electricity, and
power, not chips, is now the binding constraint.

8
00:01:10,600 --> 00:01:19,500
America's answer was to starve China of chips, and by
Musk's own account that lead is closing faster than
most people realise.

9
00:01:20,600 --> 00:01:29,500
Each side holds half of what it needs, and neither
eases off while the other accelerates, which is the
definition of a chicken game.

10
00:01:30,600 --> 00:01:39,500
And the watchers are not neutral - the man proposing
the system has been in court with one of the labs he
would be watching.

11
00:01:40,600 --> 00:01:49,500
Meanwhile that lab is worth eight hundred and fifty
billion dollars and preparing to list, which is not
the posture of an industry expecting brakes.

12
00:01:50,600 --> 00:01:57,600
So the plan is that they watch each other, a decade of
warnings ending not in rules, but in a standing
invitation to keep going.