WEBVTT

00:00.000 --> 00:04.340
these AI companies, I think what's really hard to figure out when we're doing

00:04.340 --> 00:08.920
coverage for this, and when I say we, I mean, just as a society is what is hype

00:08.920 --> 00:10.520
and what is reality.

00:10.720 --> 00:13.980
And I think these AI companies, even when negative things happened,

00:14.100 --> 00:17.060
they're still incentivized to paint them in a positive way.

00:17.280 --> 00:20.260
So, you know, that's kind of the hidden dark side here.

00:20.420 --> 00:22.240
When open AI says, oops, we hacked someone.

00:22.380 --> 00:23.940
It's like, wait, like we hacked someone.

00:24.100 --> 00:25.760
Like that's how good our models are, right?

00:25.880 --> 00:26.760
Like that's pretty cool.

00:26.760 --> 00:29.900
So it makes it really hard to actually analyze the coverage.

00:30.100 --> 00:32.980
And in this case specifically, both companies were actually

00:32.980 --> 00:35.660
incentivized by the same thing when you think about it from a PR

00:35.660 --> 00:39.340
perspective, open AI is incentivized to say, wow, like our

00:39.340 --> 00:40.520
model is just so good.

00:40.640 --> 00:41.560
Look at what it did.

00:41.780 --> 00:43.340
And it was just so out of control.

00:43.520 --> 00:45.500
It was so good that it wasn't even our fault.

00:45.880 --> 00:49.540
And then you have also hugging face who actually made a lot

00:49.540 --> 00:52.700
of very simple mistakes that could have prevented this attack,

00:52.760 --> 00:55.560
which also a human could have exploited now that we know more

00:55.560 --> 00:58.980
of the details and hugging face is also incentivized to say,

00:59.140 --> 01:01.540
oh, yeah, dude, this is like some novel crazy thing.

01:01.620 --> 01:02.640
Couldn't have been prevented.

01:02.940 --> 01:05.620
They tried to spin the entire incident as a success story at

01:05.620 --> 01:07.740
hugging face because the security team treated open

01:07.740 --> 01:10.500
AI as report with the urgency it deserved as a genuine zero

01:10.500 --> 01:12.600
day unknown to the world and moved accordingly.

01:12.960 --> 01:15.920
Left out of the post is that five days passed until open

01:15.920 --> 01:19.680
AI revealed its role on the breach hugging face disclosed and

01:19.680 --> 01:22.040
that at least another five days passed from the time open AI

01:22.040 --> 01:24.940
reported the zero days and JFrog released patches for them.