WEBVTT

00:00:01.070 --> 00:00:04.357
POSSIBLY ADD. AND THIS IS WHAT
WE WILL REWARD YOU FOR DOING.

00:00:04.440 --> 00:00:08.297
AND I THINK THAT'S WHAT'S REFLECTED
IN WHAT YOU'RE SEEING IN THAT S-1.

00:00:08.380 --> 00:00:11.257
>> COMING BACK TO YOU IN A
LITTLE BIT, THE MOMENTS HERE.

00:00:11.340 --> 00:00:14.367
SAM ALTMAN IS LIVE WITH OUR KATE
ROONEY. KATE, TAKE IT AWAY.

00:00:14.450 --> 00:00:15.827
>> HEY SCOTT, THANK YOU SO MUCH.

00:00:15.910 --> 00:00:17.537
AND SAM, IT'S GREAT TO SEE YOU.

00:00:17.620 --> 00:00:20.437
THANK YOU SO MUCH FOR HAVING
US HERE AT DEVELOPER DAY.

00:00:20.520 --> 00:00:22.337
>> THANK YOU FOR COMING TO IT.

00:00:22.420 --> 00:00:25.407
>> IT'S NICE TO HAVE AN EVENT
IN SAN FRANCISCO FOR ONCE.

00:00:25.490 --> 00:00:27.577
>> I LOVE SAN FRANCISCO.
>> YEAH. WELL, NO, IT'S BUSY

00:00:27.660 --> 00:00:31.417
EVENT. WE'VE HAD PEOPLE LINING
UP TO GET IN AND THERE'S A LOT

00:00:31.500 --> 00:00:34.517
TO TALK ABOUT. I DO WANT
TO START ON THE AI SAFETY DEBATE.

00:00:34.600 --> 00:00:37.517
THERE IS A TON TO TALK ABOUT,
BUT THE LATEST HEADLINE THAT WE

00:00:37.600 --> 00:00:39.727
HAVE CONFIRMED IS THAT OPENAI
ESSENTIALLY ABANDONED PLANS TO

00:00:39.810 --> 00:00:42.557
LAUNCH A NEW MODEL BECAUSE
OF SAFETY CONCERNS. IF YOU COULD

00:00:42.640 --> 00:00:45.527
WALK US THROUGH THAT, BUT ALSO
WHAT HAS CHANGED? YEAH. WHY DO

00:00:45.610 --> 00:00:48.797
THAT? AND IS THIS DIFFERENT THAN
WHAT YOU'VE DONE IN PREVIOUS MODELS?

00:00:48.880 --> 00:00:52.367
>> AI HAS TO BE BUILT FOR PEOPLE.
IT HAS TO BE GOOD FOR

00:00:52.450 --> 00:00:55.507
PEOPLE. PART OF THAT IS THERE
HAS TO BE SAFE. IT HAS TO BE

00:00:55.590 --> 00:00:58.577
ALWAYS UNDER HUMAN CONTROL.
IT HAS TO DO WHAT PEOPLE WANT. AND

00:00:58.660 --> 00:01:01.647
YOU WANT TO BE ABLE TO LIKE,
USE IT AND BE CONFIDENT THAT

00:01:01.730 --> 00:01:03.477
NOTHING BAD IS GOING TO HAPPEN.

00:01:03.560 --> 00:01:06.347
AS THE MODELS HAVE GOTTEN MORE
CAPABLE, THE SAFETY CASES THAT

00:01:06.430 --> 00:01:09.687
WE NOW USE HAVE TO BE HELD TO A
HIGHER AND HIGHER STANDARD, AND

00:01:09.770 --> 00:01:13.287
THE MODELS CAN DO MORE.
SO WE WILL HAVE MANY GREAT NEW MODELS

00:01:13.370 --> 00:01:15.727
TO COME. BUT WHAT WE HAVE
SAID IS THAT ALIGNMENT, SAFETY,

00:01:15.810 --> 00:01:17.827
MONITORING, SECURITY HAVE
TO STAY WAY AHEAD OF CAPABILITIES.

00:01:17.910 --> 00:01:21.427
AND WE HAVE HAD MODELS IN THE
PAST WHERE WE SAY, THIS IS NOT

00:01:21.510 --> 00:01:24.537
QUITE AT THE BAR WE WANT,
AND WE'LL MAKE ANOTHER ONE AND FIX

00:01:24.620 --> 00:01:28.237
IT AND LAUNCH IT LATER. BUT THERE'S
CLEARLY MORE ATTENTION ON THIS NOW.

00:01:28.320 --> 00:01:30.437
>> OKAY. WHAT ABOUT THIS
PARTICULAR MODEL WORRIED YOU OR

00:01:30.520 --> 00:01:33.407
FREAKED YOU OUT TO THE POINT
WHERE YOU WOULDN'T LAUNCH IT?

00:01:33.490 --> 00:01:35.577
>> I WOULDN'T OVERROTATE
ON THIS ONE THING. IT WAS LIKE,

00:01:35.660 --> 00:01:39.387
YOU KNOW, THIS MODEL WAS A
LITTLE BIT WORSE ON A FEW OF

00:01:39.470 --> 00:01:42.517
THE EVALS WE LOOK AT. THIS WAS,
THINK OF THIS IN THE ABUNDANCE

00:01:42.600 --> 00:01:45.557
OF CAUTION CATEGORY. OKAY, THERE'S
NOT A BIG SCARY THING IN THIS CASE.

00:01:45.640 --> 00:01:47.527
>> DOES THIS COUNT AS A PAUSE?

00:01:47.610 --> 00:01:50.527
AND I WONDER WHAT THE PRESIDENT
WOULD SAY IN A MOMENT LIKE THIS

00:01:50.610 --> 00:01:54.367
IF HE IS CONCERNED ABOUT SLOWING DOWN TOO
MUCH. WHAT HAVE YOUR CONVERSATIONS BEEN

00:01:54.450 --> 00:01:57.467
LIKE? AND DO YOU WORRY THAT HE
LOOKS AT A HEADLINE LIKE THIS

00:01:57.550 --> 00:01:59.507
AND SAYS, THIS COUNTS AS A PAUSE HERE?

00:01:59.590 --> 00:02:02.577
>> AGAIN, I WOULD PUT THIS IN
LIKE THE NORMAL COURSE CATEGORY

00:02:02.660 --> 00:02:05.977
OF, YOU KNOW, OFTEN WE BUILD
A MODEL, WE TEST IT, IT DOESN'T

00:02:06.060 --> 00:02:09.107
MEET OUR STANDARDS, WE CHANGE IT,
WE LAUNCH IT LATER. BUT AS

00:02:09.190 --> 00:02:13.487
WE HAVE TALKED ABOUT,
WE ARE PACING OUR PROGRESS, WHICH

00:02:13.570 --> 00:02:15.317
INCLUDES SOMETIMES NOT TRAINING

00:02:15.400 --> 00:02:21.127
THE MODEL, THE. EVERYBODY WANTS
GREAT MODELS. WELL, ACTUALLY

00:02:21.210 --> 00:02:24.357
NOT EVERYBODY. MANY PEOPLE WANT GREAT.
WE WANT GREAT MODELS,

00:02:24.440 --> 00:02:28.467
BUT WE HAVE TO DO THESE
IN A WAY WHERE WE CAN MAKE VERY

00:02:28.550 --> 00:02:31.967
CONFIDENT SAFETY CASES
AND CLAIMS SO THAT PEOPLE DON'T

00:02:32.050 --> 00:02:35.237
HAVE A BUNCH OF ANXIETY AND CAN
RELY ON US FOR THE MODELS WE'RE

00:02:35.320 --> 00:02:38.307
PUTTING OUT INTO THE WORLD.
WE ARE NOW IN A PERIOD WHERE THE

00:02:38.390 --> 00:02:41.277
MODELS ARE GETTING SO GOOD,
SO FAST, THAT WE THINK IT IS

00:02:41.360 --> 00:02:44.947
PRUDENT AND WISE TO MAKE MORE
PROGRESS ON THINGS. WE WERE

00:02:45.030 --> 00:02:47.747
TALKING ABOUT SAFETY ALIGNMENT,
MONITORING SO THAT WE CAN

00:02:47.830 --> 00:02:51.157
CONTINUE WITH PROGRESS.
BUT I THINK THIS IDEA OF PACING IS

00:02:51.240 --> 00:02:53.987
REALLY IMPORTANT. IT'S NOT JUST
LIKE, STOP IT BECAUSE, YOU KNOW,

00:02:54.070 --> 00:02:56.427
WE DON'T WANT SMARTER MODELS.
OF COURSE WE WANT BETTER MODELS.

00:02:56.510 --> 00:03:00.427
IT IS MAKE SURE THAT WE IMPROVE THE AREAS
WE NEED TO CONFIDENTLY AND SAFELY CONTINUE

00:03:00.510 --> 00:03:04.237
WITH PROGRESS. SO I. >> THINK THE PACING
DOESN'T ALWAYS MEAN SLOWING. AND I

00:03:04.320 --> 00:03:06.007
THINK THAT'S SOMETHING
THAT HAS CONFUSED PEOPLE.

00:03:06.090 --> 00:03:08.667
>> YEAH, IT MEANS DOING IT IN A
RESPONSIBLE WAY WITH GUARANTEES

00:03:08.750 --> 00:03:12.177
THAT THE WORLD CAN RELY ON.
>> HUGGING FACE. WE GOT TO TALK

00:03:12.260 --> 00:03:14.577
ABOUT THAT FOR A SECOND.
THAT WAS JUST THIS WATERSHED MOMENT,

00:03:14.660 --> 00:03:17.577
AND IT FEELS LIKE WE'VE HAD
EXAMPLES EVERY OTHER DAY SINCE

00:03:17.660 --> 00:03:21.887
THEN. DID YOU UNDERESTIMATE
AT THE TIME THE DEPTH OF THESE

00:03:21.970 --> 00:03:25.127
ISSUES? I MEAN, THERE'S BEEN REPORTS
THAT THERE ARE THOUSANDS OF THESE. HAS IT

00:03:25.210 --> 00:03:29.757
SURPRISED YOU? >> WE'RE TRYING TO I
THINK PART OF THE WAY WE WANT TO DO THIS

00:03:29.840 --> 00:03:31.797
IS TO REALLY OVERCOMMUNICATE
AND BE EXTREMELY TRANSPARENT.

00:03:31.880 --> 00:03:34.167
SO THERE'S NOTHING ELSE
THAT I'M AWARE OF OR THAT WE'RE

00:03:34.250 --> 00:03:37.137
AWARE OF THAT IS AS SERIOUS
AS THE HUGGING FACE INCIDENT. BUT

00:03:37.220 --> 00:03:41.037
WE ARE TRYING TO PUT OUT EVERY
SMALL OR OTHERWISE CASE WE CAN

00:03:41.120 --> 00:03:45.347
FIND. AND WE REALLY, AS PART
OF OUR MISSION, WANT THE WORLD TO

00:03:45.430 --> 00:03:49.247
UNDERSTAND WHAT'S WHAT THESE
MODELS ARE CAPABLE OF. ONE OF

00:03:49.330 --> 00:03:52.177
THE THINGS THAT'S SPECIAL
ABOUT OPENAI IS WE HAVE A LOT OF

00:03:52.260 --> 00:03:55.017
BREADTH. WE DO CONSUMER,
WE DO ENTERPRISE, WE DO DEVELOPERS.

00:03:55.100 --> 00:03:59.370
AND, AND PEOPLE HAVE VERY DIFFERENT WAYS
OF THINKING ABOUT WHERE WE ARE WITH A