*WhatsApp принадлежит компании Meta, которая признана экстремистской и запрещена в РФ
**https://www.youtube.com/watch?v=0pBPEN1FcFU
***https://300.ya.ru/v_Tdbs3NNO
таймкоды
00:00:00 Введение и достижения XCII
- XCII достигла значительных результатов за два с половиной года.
- Команда стала лидером в области озвучивания, создания изображений и видео.
- Модель прогнозирования Groc «420» превосходит другие ИИ и методы прогнозирования.
00:00:59 Инновации и технологии
- Запуск Gropedia, которая превзойдёт Википедию по полноте и точности.
- Создание стопроцентного обучающего кластера с графическим процессором.
- Ускоренное развитие X-AI делает компанию лидером в своей области.
00:02:28 Реорганизация компании
- Компания проходит через этапы роста и реорганизации.
- Новая структура включает четыре основные области применения: Grock Main и Voice, кодирование, изображения и видео, макроэмуляция.
00:04:03 Развитие голосового интерфейса
- Разработка модели голосового интерфейса с нуля за шесть месяцев.
- Достижение более двух миллионов тесла и API голосового агента Groc.
- Цель — создать портал для выполнения работы с помощью ИИ.
00:07:08 Модели кодирования
- Модели кодирования создают качественный код и ускоряют работу инженеров.
- ИИ понимает интуицию инженеров и может отлаживать код.
- Рекурсивное самосовершенствование кода Grog.
00:08:50 Будущее программирования
- Назначение Макроса руководителем разработки кода.
- Создание лучшей модели кодирования для всех.
- Ограничивающие факторы: вычислительные ресурсы и энергия.
00:09:45 Перспективы ИИ в программировании
- ИИ может создавать оптимизированные двоичные файлы напрямую.
- Традиционное кодирование может стать ненужным к концу года.
- Код Grok станет современным через два-три месяца.
00:10:32 Визуализация и Imagine
- Команда Imagine начала работу над визуализацией и видеогенерацией полгода назад.
- Выпущен Imagine V1, который возглавил список лидеров.
- Планируются новые релизы в этом и следующем месяцах.
00:11:10 Показатели продукта
- За полгода Imagine интегрирован во все продукты компании.
- Пользователи ежедневно создают около 50 миллионов видеороликов с помощью Imagine.
- За последние 30 дней сгенерировано 6 миллиардов изображений.
00:12:11 Цели и перспективы
- Цель Imagine — воплотить в реальность всё, что можно представить.
- Расширение возможностей моделей для создания визуальных миров, неотличимых от реальности.
- Возможность создания видеороликов продолжительностью 10–20 минут.
00:13:30 Будущее видеогенерации
- Рендеринг видео в реальном времени.
- Взаимодействие с виртуальными мирами в реальном времени.
- Лидерство в понимании и генерации видео.
00:14:36 Macro Heart
- Macro Heart создаёт цифровой эмулятор, работающий в режиме реального времени.
- Эмуляция человеческих компаний с цифровой продукцией.
- Повышение производительности через крупномасштабные вычисления.
00:17:43 Инфраструктура и API
- Работа над инфраструктурой основного продукта и API.
- Важность надёжности и безопасности.
- Решение сложных распределённых задач с беспорядочными данными.
00:18:34 Оценка качества моделей
- Привлечение экспертов для оценки качества моделей.
- Улучшение моделей в медицине, финансах и юриспруденции.
- Создание eval, полезного при обучении данным.
00:19:56 Ropedia
- Цель Ropedia — собрать воедино все человеческие знания.
- Достижение 6 миллионов статей.
- Цель — сделать Rock Five независимым от поиска информации в центре обработки данных.
00:20:45 Команда ML Infra
- Разработка программного обеспечения для обучения, логического вывода и инструментария.
- Пример: система предварительного обучения для Rock Three.
- Проблемы с масштабированием программного обеспечения при увеличении количества систем.
00:21:26 Сложности в центре обработки данных
- Переключатели, ссылки, графические процессоры выходят из строя.
- Необходимость логической работы системы при любых сбоях.
- Успешное решение проблем в период рождения сына.
00:22:22 Команда RL и логического вывода
- Обучение с подкреплением и производственный логический вывод в больших масштабах.
- Оптимизация стека: параллелизм, предварительное заполнение, декодирование.
- Устойчивость к аппаратным сбоям.
00:23:21 Команда разработчиков инструментов
- Создание платформ, фреймворков и инфраструктуры для людей и агентов.
- Платформа обработки данных о людях и внутренняя инженерная платформа.
- Важность хорошего интерфейса и полезной платформы.
00:24:13 Команда JAX
- Оптимизация сверхмасштабного обучения на графическом процессоре.
- Поддержка моделей pre-training foundation, RL и мультимодальных систем.
- Настройка стека JAX от компилятора до времени выполнения.
00:25:22 Команда kernels
- Оптимизация низкоуровневых системных задач и алгоритмов.
- Взаимодействие сотен тысяч потоков внутри графического процессора.
- Выжимание максимальной производительности из графических процессоров.
00:26:00 Компьютерная и сетевая инфраструктура
- Поддержка крупнейшего в мире суперкомпьютера в Мемфисе.
- Объединение программного и аппаратного обеспечения.
- Строительство центров обработки данных и прокладка оптоволокна.
00:28:24 Вертикальная интеграция и эффективность
- Вертикальная интеграция архитектуры, механики, электротехники.
- Достижение высокого уровня Poe и использование технологий сглаживания энергопотребления.
- Партнёрство с Tesla и другими компаниями.
00:29:04 Преимущества X.AI
- Преимущество в вычислительной мощности для успеха в ИИ.
- Быстрое подключение ИИ к сети.
- Сравнение с другими компаниями, подчёркивающее уникальность X.AI.
00:30:20 Введение и достижения компании X
- Никита, сотрудник компании X, рассказывает о поддержке клиентов.
- Семейство приложений X охватывает более миллиарда человек.
- X — важный инструмент коммуникации, где собираются влиятельные люди.
00:31:04 Успехи приложения
- Январь был самым успешным месяцем по вовлечённости.
- Февраль на пути к превышению этого показателя.
- Команда разработчиков алгоритмов достигла значительных успехов.
- Количество скачиваний с первого раза увеличивается более чем на 50% каждый месяц.
- Новые пользователи проводят в приложении на 55% больше времени в день.
00:31:59 Перестройка приложения
- Перестроены уведомления, веб-браузер и XChat.
- Количество опубликованных статей выросло в десять раз, статей Red — в семнадцать раз.
- Подписки значительно увеличились в праздничные дни.
- Годовой доход превысил миллиард долларов.
00:32:49 Новые функции и прозрачность
- Преобразование старого стека сообщений Twitter в зашифрованную систему.
- Открытие исходного кода приложения и алгоритма рекомендаций.
- Grok Chat будет с открытым исходным кодом.
00:33:46 X-Chat и X-Money
- X-Chat будет отдельным приложением с общим доступом к рабочему столу и многопользовательской работой.
- X-Money проходит закрытое бета-тестирование и скоро выйдет на внешний рынок.
00:34:39 Будущее X-App
- Цель — сделать X-App более полезным и привлекательным для пользователей.
- Ожидание более миллиарда активных пользователей в день.
00:35:20 Космические амбиции
- Объединение SpaceX и X.AI для исследования Вселенной.
- Использование солнечной энергии требует выхода за пределы Земли.
- Запуск орбитальных центров обработки данных мощностью от 100 до 200 гигаватт в год.
00:37:11 Будущее на Луне и в космосе
- Создание фабрик и спутников с ИИ на Луне.
- Использование массового двигателя для запуска спутников в дальний космос.
- Возможность достижения нескольких процентов солнечной энергии.
00:38:09 Заключение
- Мечта о массовом двигателе на Луне и самоподдерживающемся городе.
- Исследование Солнечной системы и звёздных систем.
- Встреча с инопланетянами и обнаружение древних цивилизаций.
00:38:43 Эпичное завершение презентации
- Подчёркивается невероятная скорость прогресса XAI.
- XAI конкурирует с хорошо финансируемыми компаниями, несмотря на молодой возраст.
- XAI демонстрирует успехи в области голосовой связи, создания изображений и редактирования.
00:39:38 Достижения XAI
- Меньшая команда XAI достигла значительных результатов с меньшим капиталом.
- XAI генерирует больше изображений и видео, чем все конкуренты вместе взятые.
- Интеграция с платформой X открыла эксклюзивный поток данных для обучения.
- Запуск Grokkopedia демонстрирует огромные перспективы.
00:40:31 Технические достижения
- XAI стала первой компанией с обучающим кластером в 100 тысяч часов на 100 графических процессоров.
- Планируется достижение эквивалента миллиона H100 графических процессоров.
- Реструктуризация проводится для повышения эффективности по мере роста компании.
00:41:31 Основные направления деятельности
- XAI выделила четыре основных направления: Grock Main и Voice, программирование, создание изображений и видео, макросъёмка.
- Презентация была массовым рекрутинговым мероприятием.
- Пользователи генерируют почти 50 миллионов видеороликов и около 6 миллиардов изображений в день.
- К концу года пользователи смогут создавать десятиминутные или двадцатиминутные видеоролики на основе текстовых подсказок, которые будут отображаться в режиме реального времени.
00:42:24 Революция в контенте
- Анализ любимых эпизодов телешоу «XYZ» для создания нового контента в реальном времени.
- Настраиваемые подсказки в реальном времени, адаптированные под пользователя.
- Влияние на индустрию видеоигр и Netflix.
00:43:20 Macro Hard и человеческие эмуляторы
- Разработка человеческих эмуляторов для компьютеров.
- Возможность создания цифровых компаний, подобных Apple и NVIDIA.
- Macro Hard планирует проникнуть в различные индустрии: ракеты, чипы ИИ, физика, обслуживание клиентов.
00:44:18 X и XChat
- X приносит годовой доход в миллиард долларов от подписки.
- XChat нацелен на конкуренцию с WhatsApp и Telegram.
- Стандартные функции XChat: исчезновение сообщений, блокировка скриншотов.
00:45:17 X Money
- Закрытое бета-тестирование X Money, ограниченное внешнее бета-тестирование и глобальный запуск.
- Цель X Money — стать центральным местом для всех транзакций.
- Влияние на глобальную индустрию финансовых услуг.
00:47:10 X AI и SpaceX
- Увеличение вычислений для обучения ИИ через орбитальные центры обработки данных.
- Заявка SpaceX на запуск миллиона спутников с ИИ.
- Долгосрочные планы: заводы на Луне и гигантский массовый двигатель для запуска спутников.
00:48:09 IPO SpaceX
- Привлечение капитала для запуска проекта орбитальных центров обработки данных.
- Возможность обеспечения мощности более 1000 гигаватт в год.
- Серьёзность намерений SpaceX в реализации проекта.
In this video
Timeline
Chapters
Transcript
Search transcript
Chapter 1: Elon Musk’s Opening Remarks xAI “All Hands” Meeting — xAI Accomplishments Since Inception
0:00
0 seconds
Welcome [applause] to the uh XEI all hands. We’ve got a very exciting presentation for you. We’re going to start off by recapping the incredible
0:08
8 seconds
progress that the XI team has made in uh just 2 and a half years. Uh it’s really remarkable uh in pursuit of our goal of understanding the universe. So just
0:16
16 seconds
going over our accomplishments since inception. Uh it’s important for to bear in mind that uh XI is only 2 and 1/2
0:23
23 seconds
years old, basically a toddler. uh and we’ve nonetheless achieved an incredible amount in a very short period of time.
0:30
30 seconds
So our competitors are uh 5 10 some cases 20 years old. Uh they’ve have much
0:37
37 seconds
larger teams. They started off with far more resources and yet nonetheless we have achieved number one in many arenas
0:44
44 seconds
uh in in just a few years. So uh we we’ve achieved number one in in voice uh in image and video generation. I think
0:51
51 seconds
we now at this point are actually generating more uh images and video based on the last numbers I saw than all of our competitors combined. Um we uh
1:01
1 minute, 1 second
are winning in terms of forecasting which is one of the key metrics of intelligence. Uh so our the Grock uh 420
1:08
1 minute, 8 seconds
forecasting model beat all the other AIs in forecasting. Um and we’ve talked many leaderboards. Uh we’ve got uh now a
1:15
1 minute, 15 seconds
great uh app with the with with the imagine with the core gro. We’ve made radical improvements to the X app and
1:23
1 minute, 23 seconds
we’ve launched Graipedia which is on its way to far exceeding Wikipedia and ultimately be orders of magnitude of
1:30
1 minute, 30 seconds
more comprehensive and more accurate and have more information as well as video and and and image data that simply isn’t
1:38
1 minute, 38 seconds
there on Wikipedia. So it’s uh it’s intended ultimately to be encyclopedia Galactica, a distillation of all knowledge of uh yeah all knowledge. Um
1:47
1 minute, 47 seconds
and uh we we’ve we’re uh we’re the first to achieve 100,000 uh H100 GPU training cluster and we’re now about to achieve
1:56
1 minute, 56 seconds
the first 100 I should say 1 million H100 GPU equivalents in training. So really an incredible amount of work in a
2:05
2 minutes, 5 seconds
very short period of time. And is it’s important to consider for competitiveness of any technology company what matters is not the position
2:13
2 minutes, 13 seconds
at any point in time but what is your velocity and acceleration and if you’re moving faster than anyone else uh in any
2:20
2 minutes, 20 seconds
given technology arena you will be the leader and XAI is moving faster than any other company no one’s even close so let’s go to our team uh as we uh grow as
2:30
2 minutes, 30 seconds
a company a natural thing that happens is you reorganize the company as it scales up so when you first have a startup You might have just a few dozen
2:38
2 minutes, 38 seconds
uh people and they all just chat amongst themselves. As you grow to several hundred people, you have to then add more structure just like an organism
2:46
2 minutes, 46 seconds
that grows from a single like we all just grew from a single cell um and then to a blob of cells and then you get or organ differentiation limbs. You grow a
2:55
2 minutes, 55 seconds
tail. Hopefully the tail disappears and then uh you become a baby. You go through these stages and so we’re uh
3:03
3 minutes, 3 seconds
organizing because we’ve we’ve reached a certain scale. we’re organizing the company uh to be more effective at this scale. Um now naturally when when this
3:11
3 minutes, 11 seconds
happens there there’s some people who are better suited for the early stages of a company um and and less suited for the later stages and so uh and and for for the people that that have departed.
3:22
3 minutes, 22 seconds
I’d just like to say thank you for contribution uh thank you for getting us this far and we wish you very well in your future endeavors. So now going on to uh the new structure of the company.
3:32
3 minutes, 32 seconds
Uh the company is organized uh in four main application areas. There’s there’s gro main and voice which is really the the the main gro model. That’s why it’s
3:40
3 minutes, 40 seconds
called grock main. Um then there’s a coding specific model. There’s an image and video model which is imagine and then macro hard which is intended to do
3:49
3 minutes, 49 seconds
full digital emulation of entire companies. And and then we’ve got uh the infrastructure layers. So I’d like to invite members of the team to come up and talk about each of their areas.
Chapter 2: Elon & xAI Team Give Big Update
4:03
4 minutes, 3 seconds
Hey, thanks Elon. So Grock main and voice are going to be merged into one team and you know on voice one anecdote
4:10
4 minutes, 10 seconds
is September 2024 OpenAI had this product you could talk to advanced voice mode and we had nothing no model uh of course no product. Uh we started much
4:19
4 minutes, 19 seconds
after that and in a span of few months 6 months we developed the model inhouse from scratch without a bunch of people who knew audio and had a product that was surpassing open air in 6 months.
4:29
4 minutes, 29 seconds
Fast forward six more months and now we have Groc in more than 2 million Teslas.
4:33
4 minutes, 33 seconds
We have a Grock voice agent API you can do all kinds of amazing things. In a span of one year we went from nothing to being leaders. That kind of stuff is
4:41
4 minutes, 41 seconds
only possible in a place like XCI where you have small teams committed mission focused lots of compute. Um and we really really want to keep pushing. Same
4:49
4 minutes, 49 seconds
story on the chat models. You know, we’ve always been at the forefront of uh reasoning uh starting from Grock 1.5, Grock 2, Grock 3. And we want to really
4:56
4 minutes, 56 seconds
move to a world where it’s no longer about just question answering. We want to build an everything app. So you should be able to come to it and really get done whatever you want. You know,
5:05
5 minutes, 5 seconds
ask a legal question, uh make a slide deck or or you know, solve a puzzle, stuff like that.
5:10
5 minutes, 10 seconds
Yeah. So I really think on the product side we’re really going to see a huge transformation happening in a very short period of time. Uh we’re going to see
5:18
5 minutes, 18 seconds
work the magnitude of amount of work that all uh knowledge workers are going to be able to produce increase 10fold in the next short period of a few months.
5:26
5 minutes, 26 seconds
Uh the models that we are building out are incredibly uh amazing and we have a lot on the way and we’re really excited to share that with with you all. Um, and
5:35
5 minutes, 35 seconds
on a product side, the goal is to just build that portal that allows you to accomplish all of your work. And how do we amplify everyone to achieve much much
5:43
5 minutes, 43 seconds
more than what they can accomplish alone? And we’re building that out. And it’s it’s going to be an incredibly easy to use experience that just works seamlessly.
5:51
5 minutes, 51 seconds
Uh, that being said, we are hiring and we’re looking for intelligent and smart people. This is not an easy place to work, guys. Like, this is it’s a grind,
5:58
5 minutes, 58 seconds
but we have, I guess, like interstellar ambitions, so it’s it’s not going to be easy, right? So I will say having come
6:05
6 minutes, 5 seconds
to XAI uh it has been an opportunity of a lifetime to work among really smart and really passionate people. The vibes
6:12
6 minutes, 12 seconds
here are amazing and it’s truly an environment where if you’re a smart person and you want to get done, you can get done. There isn’t like
6:20
6 minutes, 20 seconds
organizational overhead getting your way or kind of I don’t know like having to write docs and all this kind of stuff.
6:26
6 minutes, 26 seconds
You just do stuff at least at least for me. I just you know just you can do things here and that’s amazing and I invite more people to come here and just do awesome things.
6:35
6 minutes, 35 seconds
Yeah. So with with the the Grock main the sort of main um foundation model the intent is that it’s it’s genuinely
6:42
6 minutes, 42 seconds
useful in a wide range of areas. So if you’re doing uh engineering or law uh or trying or or medicine any anything uh it is useful uh to you in in in your job.
6:53
6 minutes, 53 seconds
uh that’s uh essential to understanding the universe and and making things as useful as possible like where when Grock gives you an answer that you can count on it.
7:01
7 minutes, 1 second
Absolutely. Right. All right. Thank you. Yeah. Thanks.
7:08
7 minutes, 8 seconds
Hey everybody, I’m MRO. So the world changed a lot recently in terms of coding. Um the coding models I was always complaining people were trying to
7:16
7 minutes, 16 seconds
convince me to use a coding model and I was like trusting it and I wasn’t really convinced. But as of recently the models
7:23
7 minutes, 23 seconds
they they actually produce good decent quality code. You I mean you still need to review and give feedback but you can it’s it’s easy to see how they can
7:32
7 minutes, 32 seconds
accelerate you quite a lot. So it’s not only about coding it’s like they understand your intuition like much better than before. Like now when you
7:41
7 minutes, 41 seconds
when I describe a problem I only have to phrase it like I would to an other colleague engineer who has already seen the codebase. That’s a huge change.
7:48
7 minutes, 48 seconds
before you kind of need to handhold a toddler to make a change and they don’t only write your code but they also can debug your code. So now we have I do
7:56
7 minutes, 56 seconds
like what we do like hours of Grock code running continuously to make sure that more complex uh change to the training
8:05
8 minutes, 5 seconds
system actually works in production. So, it’s easy to see for us that this is not only about accelerating us ourselves writing code and making us 10x more
8:13
8 minutes, 13 seconds
productive, but we’re really on this path for recursive self-improvement where the current generation of Grock code is training the next generation of
8:22
8 minutes, 22 seconds
croc code. And we see that this path we on exponential takeoff here. This path will continue. So, we are doubling down on coding and making coding one of the highest priority efforts in the company.
8:32
8 minutes, 32 seconds
So if you’re out there and you’re excited about coding and you’re either either very good at training modeling or you’re a really good low-level software
8:40
8 minutes, 40 seconds
engineer interesting in systems design, this is the place to work. Like we have a million H100 equivalents uh to train
8:48
8 minutes, 48 seconds
the best coding medal in the world right now. So please join us.
8:52
8 minutes, 52 seconds
Uh yeah, I’m God. Uh I work pair with macro on coding. So it become more and more obvious to us like you know over time like we are on a path to
9:00
9 minutes
singularity at least on coding. So we decided like you know have our best engineer in the company micro to lead the coding and we’ll build the best
9:08
9 minutes, 8 seconds
coding model for everyone to empower everyone to build and for me like the main like limiting factor is probably computer and energy where they can run
9:17
9 minutes, 17 seconds
the best model to support everyone to empower everyone and with spec now we are one team and we will win on the
9:24
9 minutes, 24 seconds
compute and we are winning with space compute um and also like for every engineer right so if you are like writing kernel if you’re writing
9:32
9 minutes, 32 seconds
compiler just think about like whe still worth it maybe you should join us you know for coding effort to automate yourself a little bit like to speed it
9:39
9 minutes, 39 seconds
yourself up um yeah I think it’s like really amazing year um basically what a year to to be alive and I can already
9:46
9 minutes, 46 seconds
feel the AGI feel the AI at least for coding yeah yeah I think actually things will move maybe even by the end of this year to
9:54
9 minutes, 54 seconds
where you don’t even bother do doing coding the AI just creates the binary directly um and the AI can create a much
10:01
10 minutes, 1 second
more efficient ient binary then uh can be done by any compiler. So just say create optimized binary for this particular outcome and uh and you actually bypass even traditional coding.
10:12
10 minutes, 12 seconds
There’s there’s no that that’s an intermediate step that actually uh will not be needed probably by I’d say the end of this year. Um and uh we do expect
10:22
10 minutes, 22 seconds
uh uh Gro code to be state-of-the-art in 2 to 3 months. So it’s happening very quickly.
10:32
10 minutes, 32 seconds
also do imagine so you know I mean what do we all you do right after post AJI right you probably do like digital life so that’s what we are doing here as well
10:40
10 minutes, 40 seconds
and we have the imagine team like started pretty much from scratch like six months ago we have a few people we decided we’ll to do the imaging we’ll do
10:48
10 minutes, 48 seconds
the video gen like yeah look at what we achieved today like you know like two weeks ago we released the imagine v1 we actually top the leaderboard across like
10:57
10 minutes, 57 seconds
many of them and people really loves our product love our model and we have many more releases actually this month and next month. So yeah to me there’s like a
11:06
11 minutes, 6 seconds
very high chance like we actually may build a metaverse before meta. Um yeah I will also pass to try to to talk about like you know the the metrics we have the product.
11:16
11 minutes, 16 seconds
Yeah.
11:16
11 minutes, 16 seconds
Yeah. Like like Gorang said it’s only been 6 months since we started working on imagine. We had we had no code internally for diffusion at all uh 6
11:24
11 minutes, 24 seconds
months ago and uh basically now we’ve launched imagine on every product surface that we have including seamlessly integrating into X. So you
11:31
11 minutes, 31 seconds
can open the X app right now you can long press on any image you can edit the image you can make a video out of the image. Uh we also ran a contest recently where we had some really funny
11:39
11 minutes, 39 seconds
submissions that I’m sure many of you have seen. Uh so imagine is growing extremely extremely fast. Uh and it’s because of the speed at which we iterate. Basically we do multiple
11:47
11 minutes, 47 seconds
product updates every day. We do model updates every other week and effectively what this has led to is now users are generating close to 50 million videos
11:55
11 minutes, 55 seconds
every day using imagine. Uh and just to reiterate what Elon said earlier that to the best of our knowledge that is more than every other provider combined which
12:02
12 minutes, 2 seconds
again is an astonishing place to be uh compared to where we were 6 months ago.
12:06
12 minutes, 6 seconds
Uh we are also generating 6 billion images uh in the last 30 days. Uh nano you know Google recently posted that you know 1 billion images were generated
12:15
12 minutes, 15 seconds
using nano banana in 30 days. So, you know, we’re six times that, right? Uh and and really the goal is uh it’s not like we we don’t just want to win. We
12:23
12 minutes, 23 seconds
want to win like s like over a long period of time and have sustained greatness. And so, uh the goal with Imagine is to take anything that you can, you know, imagine and turn it into
12:31
12 minutes, 31 seconds
reality. And so, that’s that’s what we’re going to, you know, that we’re going to speedrun that basically is the goal.
12:36
12 minutes, 36 seconds
Yeah. Hey, uh I’m Hen. Uh we as we uh keep scaling our model capabilities, building visual worlds that’s
12:43
12 minutes, 43 seconds
indistinguishable from reality, we’re also uh building system that unlocks much more possibility than what we have
12:50
12 minutes, 50 seconds
right now. Um they will be able to generate the videos that’s much longer than what we have right now with stories or with souls uh of your imagine. And uh
12:59
12 minutes, 59 seconds
by the end of the year, we likely will be having models that allow you to generate videos of 10 minutes or 20
13:06
13 minutes, 6 seconds
minutes in one shot without any uh intervention. You just need to give your imagination and our model, our agents
13:13
13 minutes, 13 seconds
will do it for you. And and moreover, uh those are the videos we generate and we’re also going to allow rendering
13:22
13 minutes, 22 seconds
those. We we’re already the fastest in generating the videos and we’re going to keep pushing the extreme where we’re going to render those videos in real
13:29
13 minutes, 29 seconds
time and you will be able to imagine build and interact with your own world and the world the world will respond to
13:36
13 minutes, 36 seconds
you in real time and it is exciting future that we are going to build with ourself.
13:42
13 minutes, 42 seconds
Absolutely. Uh I my prediction is that uh uh most of AI compute is going to be uh real time video understanding and
13:49
13 minutes, 49 seconds
real-time video generation and um and we expect to be the leaders in that. It’s it’s worth emphasizing these points that um you know 6 months ago we didn’t even
13:57
13 minutes, 57 seconds
have we we had basically nothing in in in or very weak in video and image generation and editing and went in 6
14:04
14 minutes, 4 seconds
months to number one spot. Um and in fact generating more more videos and images than everyone else combined.
14:09
14 minutes, 9 seconds
We’re going to do the same thing with coding and we’re going to do the same thing with macro hard and uh and I think people will be be pretty impressed with the the Gro 4.2 model that’s coming out.
14:19
14 minutes, 19 seconds
Uh that’s uh it’s a it’s a significant improvement. Um and that’s really just that that’s that’s the small version of our new model. So uh we’ll have a medium
14:28
14 minutes, 28 seconds
and a large version that are even more intelligent.
14:36
14 minutes, 36 seconds
All right. Hi everyone. I’m Toby and I work on MacroArt, the most serious of all product names. Uh so arguably giving computers to uh humans was a good idea.
14:47
14 minutes, 47 seconds
So we’re doing the same thing thing for AI. It’s kind of like inception. We’re giving computers to computers. Um so Macro is building a fully capable
14:55
14 minutes, 55 seconds
digital realtime very important human emulator. So it’s able to do anything on a computer that a human is able to do
15:02
15 minutes, 2 seconds
including using advanced tools in engineering and medicine. So there should be rocket engines fully designed
15:09
15 minutes, 9 seconds
by AI. And in a sense, it’s one of the last few remaining areas where AI is significantly worse than humans, which
15:17
15 minutes, 17 seconds
is why I think it’s one of the most exciting areas to actually innovate in and actually change change the field. Hi
15:24
15 minutes, 24 seconds
everyone. So yeah, my name is John and uh yeah, so we’re building these strong reasoning models which are now going to control our CLI. Like we’re actively
15:32
15 minutes, 32 seconds
using these every day. they are like tremendous like productivity boost to the whole team. team. I know the voice team is like killing it on that and you
15:39
15 minutes, 39 seconds
know this is the reason why we need the compute you know we need the large scale compute to run these models to boost our own productivity but um you know 80 to
15:47
15 minutes, 47 seconds
90 95% of the world uh world software has a GUI um so that’s like you know great representation and you know to
15:56
15 minutes, 56 seconds
truly make people’s lives easier uh we need to develop models that are capable of solving day-to-day tasks uh on GUI so
16:02
16 minutes, 2 seconds
macro hard you know we will emulate a company where the output is digital. Uh so this is the obvious next step for agents. Uh macro hard will enable true
16:11
16 minutes, 11 seconds
endto-end orchestration across the desktop and it will lead to immense economic prosperity. Um so yeah we’re entering an era where we need to tackle
16:19
16 minutes, 19 seconds
the hardest of tech problems but in order to solve this we need to hire the best people. So you know think of the smartest people that you’ve worked with
16:26
16 minutes, 26 seconds
and and put them forward for for a position here. And if you can’t think of anybody like go through your phone book, go for your LinkedIn. uh you’ll be
16:35
16 minutes, 35 seconds
surprised like how how big your actual network is. And they just need three properties obviously that we want to optimize for. Are they clever? Can they
16:43
16 minutes, 43 seconds
solve hard problems? And the second property is are they driven? Do they have the ambition? Do they want to win? And the third is are they a nice person?
16:50
16 minutes, 50 seconds
Like do you want to actually work with them? Um yeah, so thank you. Yeah, the mac the macro hard project is um um over time actually will probably be our most
16:59
16 minutes, 59 seconds
important uh project because uh what we’re talking about is um emulation of entire human companies. So when you look at the most valuable companies in the
17:08
17 minutes, 8 seconds
world, they are uh their their output is digital. Um so they they don’t actually uh make hardware. So it should be
17:15
17 minutes, 15 seconds
possible to completely emulate uh any company that where the output is digital. Um and this will usher in an age of prosperity likes which uh we
17:24
17 minutes, 24 seconds
could barely imagine at this point. You need imagine to imagine it. Um so this is a big this is a big deal and this is why the words macro hard are painted on
17:32
17 minutes, 32 seconds
the roof of the training cluster uh because that’s what it’s going to build. It’s also pretty funny. Yeah. Going to be a joke.
17:44
17 minutes, 44 seconds
It’s me again. You might remember me from macro hardening computer use from a long time ago, but I also actually work on core product infrastructure and API.
17:53
17 minutes, 53 seconds
In fact, this is what I’ve done for most time uh at XAI. So, anytime you use any of our products like grock.com API authentication, you go to status.x.ai.
18:03
18 minutes, 3 seconds
This is done by the core product infra team and uh a large portion of them actually sit in London and we work with Haime over there. So, we keep the lights
18:11
18 minutes, 11 seconds
on at peak hour, 400 p.m. every day. Uh, we get paged at night when stuff goes down. Also, thank you to anyone in Palo
18:19
18 minutes, 19 seconds
Alto getting paged. Um, there’s really important work, reliability, security, core product infrastructure. So, if you actually like if you’re really
18:26
18 minutes, 26 seconds
interested in solving difficult distributed problems with like messy data, this is the team to join.
18:34
18 minutes, 34 seconds
Uh, hey everyone, my name is Diego. Um yeah so I think one one of the main bottlenecks in this next year uh for these models is going to be very high quality evals and training data and uh
18:42
18 minutes, 42 seconds
one of the ways we solve that is by taking the world’s foremost experts in these domains u bringing them here and um uh having them evaluate the model. Uh we do this for domains like medicine,
18:51
18 minutes, 51 seconds
finance, uh law, we have voice actors, we have video editors um who contribute daily um to making gro. Um and uh yeah
18:58
18 minutes, 58 seconds
we’re going to be continuing to work on very high quality u over the next few months. Um we have some exciting stuff in um you know the the frontier of uh
19:05
19 minutes, 5 seconds
useful tasks in finance and law. Um you know we’re trying to build evals that are are are useful and training data that that represents uh useful work um
19:14
19 minutes, 14 seconds
and and not necessarily proxies of intelligence where I think a lot of the open source evals do today. And yeah.
19:20
19 minutes, 20 seconds
Yeah. I’d like to say like um we’re um we’re shifting from using these sort of uh common uh internet evals which I
19:28
19 minutes, 28 seconds
think are actually not a real indicator of usefulness to uh having uh expert tutors in each domain um to every domain
19:36
19 minutes, 36 seconds
of uh engineering, medicine, law, whatever the case may be. Um and the the the actual eval is does the expert in that arena or does our group of experts
19:45
19 minutes, 45 seconds
in that arena human experts uh agree that uh Groank is extremely useful um and that the results are correct? That’s the that’s actually the only eval that really matters.
19:53
19 minutes, 53 seconds
Yeah, exactly. Um in you you you’ll see this in in Gro 420. Um but we made some improvements because of that type of data in in truth seeeking and and kind
20:02
20 minutes, 2 seconds
of minimizing political bias. The responses are are much more cogent. Um yeah, that that’s exciting. Uh and we are also working on Gracipedia. So the
20:10
20 minutes, 10 seconds
uh the goal of Rockedia is to create a distillation of all human knowledge. Um I kind of like to think of this as like a modern day version of the Library of Alexandria. Um and in the quest to build
20:18
20 minutes, 18 seconds
Encyclopedia Galactica, um which it will one day be called. Uh we’ve gone from essentially having nothing to around 6 million um articles. Um for context,
20:27
20 minutes, 27 seconds
Wikipedia is around 7 million English articles. Uh and uh yeah, we’re improving on hallucination. Um uh and
20:34
20 minutes, 34 seconds
our our our goal is essentially for rock 5 to not have to search out of the data center. Um so yeah
20:45
20 minutes, 45 seconds
I guess so in the ML infra team uh we are building the training inference and tooling team this tooling software for the company. So to give you an example
20:54
20 minutes, 54 seconds
when we were training Grock 3 we built the pre-training framework for this and it these are some some of the coolest system in my opinion that you can build
21:02
21 minutes, 2 seconds
as a software engineer. So it’s like we have 100k H100s at the time and they were just delivered and we didn’t quite
21:09
21 minutes, 9 seconds
have the software. We thought we’d have the software, but then at 30k scale, we we realized actually the software is not quite working. And it took a a major
21:19
21 minutes, 19 seconds
almost I would say halfway rewrite of the software because there’s so much going on in a data center that you can’t actually account for. Um, switches are
21:27
21 minutes, 27 seconds
switches are flapping, uh, links are flapping, switches are going down, GPUs are just burning through, you have numeric issues, and it’s a system where
21:36
21 minutes, 36 seconds
you want really 100k H100s to behave in log step. So, a training step is like 5 seconds, and you’re going 5 seconds in log, but during that 5 seconds,
21:45
21 minutes, 45 seconds
everything can happen. So, you need to write a system that makes progress despite all these things that can happen in the environment. And we did this successfully and it was one of the
21:53
21 minutes, 53 seconds
coolest times in my life where the the system was actually running at it was running at the same time my son was born. So that was extra excitement. Um
22:00
22 minutes
but these problems like you don’t find anywhere else like nobody has this kind of compute and also nobody has this kind
22:07
22 minutes, 7 seconds
of talent density. So at the time to give you a perspective we were like in overall team in pre-training we were probably like 15 people out of that
22:15
22 minutes, 15 seconds
maybe like seven people were working on the actual training system and we we still maintain that talent talent density in the team. So if you’re
22:23
22 minutes, 23 seconds
interested in working on these problems and you don’t want to be just like part of a bigger organization where you’re one of like a thousand people working on this then this is the place like we are
22:32
22 minutes, 32 seconds
still a very small team with me is Leon Min from the RL and inference team.
22:36
22 minutes, 36 seconds
Hi I’m Lemie. So at our team we run a reinforcement learning training job and production inference at large scale on the earth and probably soon in space uh
22:45
22 minutes, 45 seconds
and we are kind of already uh design a lot of thing uh to to make it more resilient and scalable. So we’re
22:52
22 minutes, 52 seconds
building a system to scale from 100k chips to millions of chips and uh we optimize every aspect of the stack like
23:00
23 minutes
parallelism prefill decode and make resilient to every known and unknown hardware failure. Um so if if you are
23:07
23 minutes, 7 seconds
system hackers obsessed with extreme performance and reliability so here is you will find uh the most interesting
23:14
23 minutes, 14 seconds
problems to work with and I think actually like very similar to all kind of thing like you it’s it’s very
23:22
23 minutes, 22 seconds
important for you to first see the problem and then you will develop the solutions that no one else can develop before I’ll hand over to the tooling team.
23:30
23 minutes, 30 seconds
Hello uh I’m Ashtip from the tooling team. uh every software uh needs to have a great interface to be able to make it
23:38
23 minutes, 38 seconds
useful. So as the tooling team we are responsible for building the platforms, frameworks and infrastructure which is required for humans as well as agents to
23:46
23 minutes, 46 seconds
be able to use our products. Uh we started by building out the human data platform. This is a place where we collect all of our human data and eventually expand it out to build our
23:55
23 minutes, 55 seconds
internal engineering platform through which we basically run deployments uh run evaluations or like look at like
24:02
24 minutes, 2 seconds
what training results exist. So if you really care about building a good interface or providing a really useful framework for researchers for agents as
24:11
24 minutes, 11 seconds
well as our tutors then you should definitely join our team. So hi everyone I’m Yulom from the Jax team. So now Jax at XAI is a really small team with a
24:20
24 minutes, 20 seconds
couple of engineers that working on Jax GPU to optimize our uh ultra large scale GPU training. So you can imagine that
24:28
24 minutes, 28 seconds
training at scale can be very complicated. Even you run hello world at scale it can be complicated right. So um then we are actually um responsible for
24:37
24 minutes, 37 seconds
supporting the entire companies from uh pre-training foundation models RLS and also multimodel to scale um things um to
24:46
24 minutes, 46 seconds
from uh first from 10k 100k then probably 1 million h100 eent equivalent
24:53
24 minutes, 53 seconds
uh GPU scale and uh we to implement a lot of you know practic practical optimizations we have to um uh customize
25:01
25 minutes, 1 second
the entire jack stack from compiler better at runtimes and there will be a lot of interesting problems. Um and also
25:08
25 minutes, 8 seconds
if you really want to uh you know u obsessed on optimizing uh the entire stack at scale that we are probably the
25:16
25 minutes, 16 seconds
best place to go uh because you know we really have very large scale uh GPU clusters and we have a lot of interesting problems to work with. Hey, I’m pranul from the kernels team.
25:25
25 minutes, 25 seconds
Basically the kernel team sits at the very bottom of our training and serving stack. Our code runs inside the million equivalent GPUs that we have. And if you
25:32
25 minutes, 32 seconds
look inside the GPU, there’s hundreds of thousands of threads. And these threads are trying to talk to each other to multiply matrices, compute attention scores, and some of them even talk to
25:41
25 minutes, 41 seconds
the million other GPUs that we have. And this is the low-level system that we have. And we like optimizing every single microcond in this. And we care deeply about squeezing every last drop
25:49
25 minutes, 49 seconds
of performance from these GPUs. So if you like this low-level systems problems, algorithms, please join us.
Chapter 3: Live Tour Of xAI’s ‘Macrohard’ AI Training Supercluster In Memphis — It’s INSANE
26:00
26 minutes
Let’s see. Now we’ll try to bring in uh Hiner and and Spencer who are actually uh at our supercomput cluster in Memphis. Hey Hiner.
26:09
26 minutes, 9 seconds
Hey.
26:12
26 minutes, 12 seconds
Hey. I’m I’m Her from the computer network infrastructure team. Uh we are mainly based in Paral, but today we’re here in Memphis in the super computer.
26:21
26 minutes, 21 seconds
So the data center here Memphis the largest compute cluster on the planet and it is still growing. Uh our job is
26:28
26 minutes, 28 seconds
to keep all this computer up and running next door and the serpent which used to work well ingredients.
26:37
26 minutes, 37 seconds
Actually just put the mic really close to your mouth cuz you’re the ambient noise is high. Ah it’s getting a lot back.
26:44
26 minutes, 44 seconds
So I was saying keep the computer up and running trying the next model of drug and serve AI so users. So for GPUs to work well uh a lot of ingredients have
26:53
26 minutes, 53 seconds
to come together mainly software and hardware. So there’s all these JB CPUs, nicks, switches in hundreds of thousands
27:00
27 minutes
of operating systems running as one big supercomputer and um what we need is folks who really understand the really understand and really understand how
27:07
27 minutes, 7 seconds
computers work on a deep level. That is you reach out and I’m handing over to Dan. All right.
27:13
27 minutes, 13 seconds
So we have 300,000 GB 300 platform to use here today. Still growing, still building 847
27:22
27 minutes, 22 seconds
miles of fiber per data hall. 12 data halls. If you want to be part of the world’s largest supercomput, come join us. All right.
27:29
27 minutes, 29 seconds
So, it’s it’s quite marvelous what we’ve been able to do in less than one year’s time here. And uh we have once we’re completely finished, we’ll have north of a gigawatt of power online and running.
27:40
27 minutes, 40 seconds
We’ll have the largest Tesla mega pack uh system in the world, larger than Hawaii or South Australia. Um and Zach is really quickly going to talk a little
27:48
27 minutes, 48 seconds
bit about actually constructing the data center. So, behind me you can see data hall 11. So, one of the most incredible things about what we’re doing here at
27:55
27 minutes, 55 seconds
MacroArts, how fast we do it, right? So, like they were saying before, over 850 miles of fiber at every single data
28:02
28 minutes, 2 seconds
hall, over 27,000 use um and over a 200,000 connections. So, all of this that you can see behind me was put up in
28:10
28 minutes, 10 seconds
less than 6 weeks. We do that over and over and over again. We massively parallelize it. It’s pretty much the most complex and consistent type of
28:19
28 minutes, 19 seconds
engineering design and construction project you possibly imagine. So, come join us. Yes.
28:25
28 minutes, 25 seconds
You know, the the other really awesome thing about this is that everything is completely vertically integrated within this team. Um, from architecture, mechanical, electrical, trucks, all the
28:33
28 minutes, 33 seconds
disciplines. Um, and we also care a lot about efficiency while we’re designing all of this too. So, it’s not just about getting the most compute online the
28:41
28 minutes, 41 seconds
fastest, but also achieving the highest POE in the industry um of using as much power smoothing technology as we can and
28:49
28 minutes, 49 seconds
being really good partners in the community here in Memphis um with the Tesla MAP packs and that we have going.
28:56
28 minutes, 56 seconds
You can check them out XAI Memphis. Back to you.
29:00
29 minutes
All right. Thank you. [applause] All right. So, that was live live from the front lines in Memphis. Um so
29:08
29 minutes, 8 seconds
fundamental to u any AI company’s success is the compute advantage and uh what we’ve demonstrated over and over again is that XAI can actually deploy uh more AI compute faster than anyone else.
29:20
29 minutes, 20 seconds
Um and actually as Jensen Hong of CEO of Nvidia has said many times uh in interviews there is no one faster at
29:27
29 minutes, 27 seconds
getting AI compute online than XAI. So congratulations guys. [applause] Yeah this is what it looks like. So
29:36
29 minutes, 36 seconds
that’s uh really phase one which is uh 330,000 uh grace blackwells uh with macro hot written on the building that’s
29:44
29 minutes, 44 seconds
uh unimag it actually is on the roof of the building. Um and then uh macro hotter will be the building that you can see which has got the macro hotter with
29:52
29 minutes, 52 seconds
the rockets on it. Um and that will be another 220,000 uh GB300s. Um so all of this will be training uh the models that
30:01
30 minutes, 1 second
you uh that you experience. So the it it’s uh absolutely fundamental obviously to have large scale training compute in order to get the best models. Yeah, I’m
30:09
30 minutes, 9 seconds
sort of reminded of the the Jose meme where you see uh one guy digging and there’s like seven people watching. Um and one of the big differences between
30:18
30 minutes, 18 seconds
XAI and and other companies is we are actually Jose.
Chapter 4: xAI’s Secret Weapon: The X Platform — Nikita Explains
30:20
30 minutes, 20 seconds
Hello. All right. I’m Nikita. You might know me as a part-time ship poster, full-time customer support for X. Um,
30:28
30 minutes, 28 seconds
so, uh, we’re now reaching over a billion people across our family of apps. Um, every time news breaks, it
30:35
30 minutes, 35 seconds
just becomes evident that this is the most important communication tool of our time. Uh, it’s where the the the most influential people, uh, come convene.
30:44
30 minutes, 44 seconds
Uh, it’s where truth is crystallized.
30:46
30 minutes, 46 seconds
Everything is downstream of X. The reason they say this is going to hit Facebook in a week because it happens here. Um, and uh, I think we’re only beginning to realize its full potential.
30:57
30 minutes, 57 seconds
Uh, we had a remarkable year for the app. Um, we rolled up our sleeves and got a ton done. Uh, January was our
31:05
31 minutes, 5 seconds
biggest month ever for the app, uh, in terms of engagement. Uh, and then February is on track to beat that. Uh, much of the credit lies with the
31:14
31 minutes, 14 seconds
algorithm team. Uh, they’ve been putting in crazy hours uh, and it’s clearly paying off, but there’s still a huge amount of work to be done. Um on the top
31:23
31 minutes, 23 seconds
of funnel side, uh first-time downloads are up over 50% uh every month. Um and we’re exhibiting right now like basically the growth rates of an early
31:31
31 minutes, 31 seconds
stage uh consumer product. Um we also made a ton of headway in solving one of the like 20-year-old problems of the app, which was ramping up new users. Uh
31:41
31 minutes, 41 seconds
new users are now spending 55% more time per day in the app uh than they were 6 months ago. Um, and on on the core
31:50
31 minutes, 50 seconds
product side, we’re we’re hitting our stride, too. Um, not only did we rebuild the algorithm, we rebuilt our onboarding flows, and we’re seeing double-digit
31:57
31 minutes, 57 seconds
increases on all our key metrics. We rebuilt notifications, our web browser, uh, X chat. Basically, every surface of
32:05
32 minutes, 5 seconds
the app has been rebuilt to be better than ever. Um, and it’s clear that if we’re focused, uh, we can move mountains and evolve this platform. Uh just last
32:14
32 minutes, 14 seconds
month we did a little push on uh articles. Um and uh articles published are up 10x. Uh articles read are up 17x.
32:24
32 minutes, 24 seconds
Um and uh on all other fronts like on over the holidays uh we did a big push on subscriptions. We just crossed a
32:32
32 minutes, 32 seconds
billion dollars in ARR there. Um I I think with the X app, you know, the there’s very few unknowns like the path
32:39
32 minutes, 39 seconds
for us to win and become uh you know the number one app in the world. Uh we’re it’s it’s we we know what to do. The
32:47
32 minutes, 47 seconds
ball’s in our court. Uh it’s it’s for us to win and it’s just a matter of us executing.
32:52
32 minutes, 52 seconds
Yep. And uh yeah, [applause] so we’ve evolved the what used to be the uh
Chapter 5: Elon On X Money, X Chat, Future Goals
33:01
33 minutes, 1 second
old Twitter DM stack, which was unencrypted, basically just text, uh to a fully encrypted uh messaging system
33:08
33 minutes, 8 seconds
that it allows you to do uh audio and video calls. Um has uh you know, all the things you’d want from any messaging app, the disingering me messages, screen
33:17
33 minutes, 17 seconds
screenshot blocks, like there’s a whole all the features that you’d want want in an app. Um we and uh we will be open sourcing the code for this in the next
33:24
33 minutes, 24 seconds
few months as we’re open sourcing the the recommendation algorithm code so people can actually see what we’re doing. Uh nothing beats nothing beats uh
33:33
33 minutes, 33 seconds
transparency for for believing in in a company. So we’re we’re the going to be the only recommendation algorithm uh
33:41
33 minutes, 41 seconds
that actually open sources so you can see what it what it does and how it’s evolving. Um with with Gro Chat it will also be uh open source so you can
33:49
33 minutes, 49 seconds
actually see if there are any vulnerabilities. There will be no hooks for advertising or anything else like that in uh in in Groch chat which is really intended to be a generalized
33:57
33 minutes, 57 seconds
communication system. Um and in the next few months we’ll be releasing a standalone uh uh X chat app. So if you just want to do messaging you can just
34:06
34 minutes, 6 seconds
you can do that. You don’t you don’t have to go to the the core product. Um and it will have desk uh desktop sharing um and u multi-user. So you can do you can do video calls with lots of people.
34:16
34 minutes, 16 seconds
It’s really intended to be a a fully functional um communication system uh with with Xhat. Uh for X Money, we’re uh
34:25
34 minutes, 25 seconds
we’ve actually had XMoney uh live in closed beta within the company. Um and we expect in the next uh month or two uh
34:32
34 minutes, 32 seconds
to go to uh a limited uh external beta and then to go worldwide to uh all X users. And this is really intended to be
34:41
34 minutes, 41 seconds
the place where all the money is the this the central source of of all uh monetary transactions. So it’s it’s a it’s really going to be a gamecher. Um
34:50
34 minutes, 50 seconds
and uh the reason we say 1 billion users is actually over a billion users is that uh while our monthly users are on
34:56
34 minutes, 56 seconds
average around 600 million, the uh number of people who have the X app installed is well over a billion. It’s just that most people only occasionally
35:05
35 minutes, 5 seconds
come to the X app when there’s some major world event. But as we give people more reasons to use the the X app,
35:12
35 minutes, 12 seconds
whether it’s for communications uh for uh rock or for uh X money, whatever the case may be, uh we we want it to be such
35:21
35 minutes, 21 seconds
that if you wanted to, you could live your life on the X app. And as you make it more and more useful, uh we’ll obviously give people reasons, compelling reasons to use the app every
35:29
35 minutes, 29 seconds
day. Um and have my expectation is well over a billion daily active users. Now um in order to understand the universe
Chapter 6: Elon Explains SpaceX & xAI Joining — “Exploring The Universe” & Moonbase Alpha
35:38
35 minutes, 38 seconds
you must explore the universe. There’s only so much you can learn from uh from just being on earth uh with telescopes and and colliders on earth. Ultimately
35:46
35 minutes, 46 seconds
you have to go out there and you have to explore the universe to understand it.
35:49
35 minutes, 49 seconds
And that’s the motivation uh behind the combination of SpaceX and XAI um is to accelerate humanity’s future in
35:58
35 minutes, 58 seconds
understanding the universe and extending the light of consciousness to the stars.
36:02
36 minutes, 2 seconds
So in the grand scheme of things, when you look at how much energy Earth is actually using for civilization, we’re only right now using call it roughly 1%
36:11
36 minutes, 11 seconds
of the potential energy of Earth. Um and if we wanted to use even a millionth of the sun’s energy, that would be roughly
36:18
36 minutes, 18 seconds
a million times more energy than civilization currently uses. The only way to access that that energy, the energy of the sun, is to extend beyond
36:26
36 minutes, 26 seconds
Earth. Earth is really a tiny tiny dust moat in in a vast darkness. uh the sun is 99.8% of all mass in the solar
36:35
36 minutes, 35 seconds
system. So you you have to expand beyond the tiny dust moat that is earth uh to to make uh any significant dent in using
36:43
36 minutes, 43 seconds
the sun’s energy. Like I said, you’d have to expand roughly a million times just to get to 1 millionth of our of our sun’s energy and then going beyond that
36:51
36 minutes, 51 seconds
exploring ex extending to the galaxy and maybe someday even to other galaxies. So the uh the the the next step beyond
37:00
37 minutes
Earth data centers is our our Earth orbital data centers. Um and we’ll be launching with SpaceX orbital data
37:07
37 minutes, 7 seconds
centers at the 100 to 200 gawatt per year level. Not cumulative, I mean per year. And ultimately we see a path to
37:15
37 minutes, 15 seconds
maybe launching as much as a terowatt per year of compute from Earth. But what if you want to go beyond a mere terawatt
37:23
37 minutes, 23 seconds
per year? In order to do that you have to go to the moon. So, by having factories on the moon building AI satellites and having a mass driver, which is the kind of thing you really
37:31
37 minutes, 31 seconds
only learn about in or read about in science fiction, but we’re going to make it real. We’re actually going to have a mass driver on the moon. And if you do
37:39
37 minutes, 39 seconds
that, you can go several orders of magnitude greater. You can go to a,000 gawatt or more per year. Um, and
37:46
37 minutes, 46 seconds
ultimately get to maybe a millionth and then a thousandth and maybe even a few% of the sun’s energy. It it’s difficult
37:54
37 minutes, 54 seconds
to imagine what an intelligence of that scale would think about, but it’s going to be incredibly exciting to see it happen. I really want to see the mass
38:01
38 minutes, 1 second
driver on the moon that is uh shooting AI satellites into deep space. It’s going like just one after the other. Um,
38:09
38 minutes, 9 seconds
I can’t imagine anything more epic than a mass driver on the moon and a self-sustaining city on the moon and then going beyond the moon to Mars. Uh,
38:17
38 minutes, 17 seconds
going throughout our solar system and ultimately uh going being out there among the stars and visiting all these star systems. Maybe we’ll meet aliens.
38:25
38 minutes, 25 seconds
Uh maybe we’ll meet see some civilizations that lasted for millions of years and we’ll find the remnants of ancient alien civilizations. But the
38:33
38 minutes, 33 seconds
only way we’re going to do do that is if we go out there and we explore. And this is the path to making it happen. Thank you.
38:40
38 minutes, 40 seconds
[applause]
Chapter 7: Recap & Key Takeaways
38:43
38 minutes, 43 seconds
Wow. Epic end to a great presentation.
38:46
38 minutes, 46 seconds
Yes, it’s done. If that’s all you wanted to see, catch you later. I’m going to share a few of my key takeaways and a bit of a recap. First of all, the incredible rate of progress. This is
38:54
38 minutes, 54 seconds
really important, not just for people who are interested in the progress of XAI themselves, but any of the Muskled engineering efforts out there. Tesla,
39:02
39 minutes, 2 seconds
SpaceX, XAI, X, Neurolink, a Boring Company, you name it. Despite being just 2 and 1/2 years old, competing with
39:09
39 minutes, 9 seconds
companies very wellfunded, deep pocketed, gigantic companies, some of them 5, 10, 20 years old, XAI has now
39:18
39 minutes, 18 seconds
equaled in terms of capabilities, if not exceeded in many areas, including voice, image generation, and image editing.
39:25
39 minutes, 25 seconds
Grock 4.2 and other variations at the top of many of the intelligence leaderboards. and Grock 420 now absolutely crushing it with forecasting
39:34
39 minutes, 34 seconds
including by the way making money in the stock market crushing all other AI models. This has been achieved with a much smaller team arriving late to the party with far less capital required.
39:46
39 minutes, 46 seconds
This matters. We also heard that as of today XAI via some of the Gro products now generating more images and video
39:55
39 minutes, 55 seconds
than all competitors combined. A phenomenal but unsurprising result.
40:01
40 minutes, 1 second
Recall believe well under a year ago, XAI acquired the X platform. Not only unlocking a massive exclusive data
40:10
40 minutes, 10 seconds
stream, all the posts on X, text, video, images for training, which is incredible, but also a massive platform to start integrating Grock, massively
40:17
40 minutes, 17 seconds
accelerating user growth. The X integration was a master stroke. They’ve also launched Groipedia, which already showing an enormous amount of promise.
40:25
40 minutes, 25 seconds
Compared to the burning dumpster fire that is a left-hearted, completely deranged, entirely captured Wikipedia, it’s only a matter of time before
40:33
40 minutes, 33 seconds
Groipedia transitions into Encyclopedia Galactica. And nobody even remembers where Gracipedia’s original name had
40:41
40 minutes, 41 seconds
come from. Another fairly direct flex, XAI, the first company with a 100,000 H100 GPU training cluster, which went up
40:49
40 minutes, 49 seconds
in record time. Jensen Wong, Nvidia CEO, on the record say no one else could have done this as quickly as XAI did. He didn’t, by the way, need to say that.
40:58
40 minutes, 58 seconds
Probably offended every single other company buying H100s from Nvidia. But the facts are the facts. They weren’t done flexing, though. XAI is about to
41:05
41 minutes, 5 seconds
become the first company with the equivalent of 1 million H100 GPUs in training. Unbelievable.
41:13
41 minutes, 13 seconds
That’s a 10x from the monumental milestone we just discussed of 100,000, which was a first. and not just the first but went up in record time. Some
41:21
41 minutes, 21 seconds
people recently concerned about a few departures from some of the XAI founding members, early team members and so on.
41:27
41 minutes, 27 seconds
Mus pointing out the company’s growing at a phenomenal rate. They’ve done some recent restructuring just so that as the company scales they are more effective
41:35
41 minutes, 35 seconds
at their new size. XAI broke down the four main areas of the company. Grock main and voice that’s one. Coding two,
41:42
41 minutes, 42 seconds
imagine eg images and video generation three. And macroart four. And at this point in time, if you hadn’t already noticed, this entire presentation was a
41:50
41 minutes, 50 seconds
massive recruiting drive. Please join us. Come join us. If you like working on hard problems, big impact, come join us.
41:57
41 minutes, 57 seconds
Join the team. We also heard some astonishing numbers. Users are currently generating almost 50 million videos per day and around 6 billion images per day.
42:09
42 minutes, 9 seconds
Unbelievable. They also expect and this is really big that by end of year users will be able to generate 10 to 20 minute
42:18
42 minutes, 18 seconds
videos from text prompts and in the future these videos will be rendered in real time. Now think about the implications. Hey Grock, my favorite TV
42:25
42 minutes, 25 seconds
show is XYZ. My favorite episodes from TV show are XYZ. Analyze why I like these episodes in particular and then create me some new content based on the
42:34
42 minutes, 34 seconds
themes, the topics, the characters I like in this particular style. Bam. Done in real time. This is going to be a revolution for content. Real time on
42:43
42 minutes, 43 seconds
demand, completely custom prompts tailored exactly to you, rendered in real time. Does this mean that Netflix
42:50
42 minutes, 50 seconds
is about to get blockbustered? Something to ponder. This will also have massive implications for the video game industry, but we’ll save that conversation for another day. After
42:59
42 minutes, 59 seconds
this, Musk pointed out that he believes in the future most AI compute will be used for realtime video understanding
43:06
43 minutes, 6 seconds
and real time video generation. So let that sink in, especially if anyone involved in the entertainment industry
43:13
43 minutes, 13 seconds
is listening. Then onto Macro hard, likely to be the most important, impactful, and lucrative project. The
43:20
43 minutes, 20 seconds
goal of macro hard is essentially developing human emulators that can use computers. Any company whose primary or sole output today is digital, which by
43:28
43 minutes, 28 seconds
the way is most companies, and absolutely the vast majority of value in terms of market capitalization is companies whose entire output is
43:35
43 minutes, 35 seconds
digital. Electrons. Think about all the tech companies, you know, the largest companies in the world. A human emulator can do all the work of a human on a computer except doesn’t need to take a
43:43
43 minutes, 43 seconds
break. Can work 24 hours a day, 7 days a week without any time off. Macroheart will be creating entire digital companies. And an important
43:51
43 minutes, 51 seconds
clarification cuz a lot of people intuitive will think, oh, you know, Apple’s safe cuz they make phones and computers. Nvidia is safe cuz they make
43:58
43 minutes, 58 seconds
AI training GPUs. No, actually, as Musk pointed out quite recently, Apple doesn’t make hardware. They design
44:05
44 minutes, 5 seconds
hardware. And then they FTP or email, whatever terminology you want to use, they send the plans for their design to Taiwan or to China or to pick your
44:13
44 minutes, 13 seconds
overseas manufacturing location. That’s who actually makes the hardware. Same is true of Nvidia. They design the chips.
44:19
44 minutes, 19 seconds
But Nvidia does not manufacture anything. They send their plans to Taiwan. They’re manufactured by third parties. Macrohart in theory, if
44:27
44 minutes, 27 seconds
successful, could replicate a company like Apple, Nvidia, Microsoft, Google. The output of most companies is digital.
44:35
44 minutes, 35 seconds
In the future, you can count on it.
44:37
44 minutes, 37 seconds
Macroh hard will have a company specifically involved in designing rockets, another AI chips, another doing physics, customer service. You name it,
44:45
44 minutes, 45 seconds
the industry, the technology, whatever it is, you can guarantee MacroArt will penetrate that market. Again, there are
44:53
44 minutes, 53 seconds
very strong recruiting efforts specifically around the MacroArt project. Come join us. Join our team.
44:58
44 minutes, 58 seconds
Another cool disclosure, X absolutely crushing it now. $1 billion in annual recurring revenue from subscriptions.
45:06
45 minutes, 6 seconds
The Xplatform itself, well, in excess of 1 billion users coming soon, not just Macroard, but a standalone X chat app.
45:14
45 minutes, 14 seconds
In other words, they’re going after the likes of WhatsApp, Telegram, blah blah blah. Importantly, with all the standard features you’d expect, including disappearing messages, the blocking of
45:23
45 minutes, 23 seconds
screenshots. So, yes, no doubt there’ll be plenty of titties being transmitted on the X Chat standalone app in the future. And this is important. This app
45:30
45 minutes, 30 seconds
will be distinct from the X platform. So if you just want to use X chat, which they’re obviously going to try and make the best chat app, period, you don’t
45:37
45 minutes, 37 seconds
even need to be an X user. And then the big one, X Money, currently in closed beta within the company soon, a limited external beta and then a global launch.
45:47
45 minutes, 47 seconds
Bit of a full circle moment for those who don’t know your history about Musk.
45:50
45 minutes, 50 seconds
Early days after founding Zip 2, selling the company, making a bit of spare change. The guys working in online banking, trying to digitize the whole
45:58
45 minutes, 58 seconds
system. Eventually, two companies emerge, X.com and PayPal, continuing with the name PayPal. Must vision at the time, online digital banking services
46:06
46 minutes, 6 seconds
for everything. And here we are a few decades later. PayPal, I believe 99, was founded, sold to eBay, I think a couple
46:14
46 minutes, 14 seconds
years after that. A big moment and a massive innovation. Musk stating quite clearly the goal is for X money to be the place where all the money is the
46:23
46 minutes, 23 seconds
central place for all transactions. Now in case you are unaware the global financial services industry today is
46:30
46 minutes, 30 seconds
Jucking Gantic. I just want to explain a few things when he’s talking about where all the money is. The central place for all transactions. We’re not just talking
46:38
46 minutes, 38 seconds
about savings accounts, transactions accounts. No, no, no. We’re talking about mortgages, business loans, construction loans, lines of credit,
46:46
46 minutes, 46 seconds
owning stocks, shitcoins, crypto, you name it. All the money. And boy, is this industry ripe for disruption. X money is going to print money for the company.
46:56
46 minutes, 56 seconds
And importantly, the big final message explaining why XAI and SpaceX are now one. In case you missed it, SpaceX
47:04
47 minutes, 4 seconds
recently acquired XAI, which again previously acquired X. So, there’s a lot of X’s in this entity. And the short answer to this is the only way for
47:13
47 minutes, 13 seconds
massive increases in the amount of AI training compute available and actually able to be turned on eg powered requires getting off Earth. Hence orbital AI data
47:22
47 minutes, 22 seconds
centers. I continue to call this project Starink at least until somebody comes up with a better name. Just to be clear, they are serious about this. SpaceX has
47:30
47 minutes, 30 seconds
already filed an application with the FCC to launch a million AI satellites for both training and inference. A million. And that’s for starters. Mus
47:38
47 minutes, 38 seconds
says ultimately annual launches from Earth for orbital data centers in space could be somewhere in the vicinity of 200 to 300 gawatt per year which is
47:46
47 minutes, 46 seconds
ridiculous but in order to get beyond that [laughter] I mean we are in the best timeline dude seriously longer term they want to have factories on the moon building
47:54
47 minutes, 54 seconds
satellites and a gigantic mass driver and not to get like too nerdy here but the purpose of the mass driver is to basically launch the AI satellites into
48:02
48 minutes, 2 seconds
orbit without needing a rocket launch but it’s basically a gigantic rail gun to shoot satellites into orbit on the moon. And they’re serious about this and
48:10
48 minutes, 10 seconds
this could get them to over 1,000 gawatt per year. And again, I reiterate this is a serious endeavor. In fact, this is
48:17
48 minutes, 17 seconds
almost the entire reason that SpaceX will be IPOing this year is to raise capital to start this truly insane
48:23
48 minutes, 23 seconds
project. What the actual Want more content, early access, bunch of perks?
Chapter 8: Want More Content? Early Access? Tesla Valuation Model?
48:29
48 minutes, 29 seconds
Click the links in the pin comment. AG1 has been part of my daily health protocol for almost half a decade. It’s packed full of vitamins and minerals and
Chapter 9: Drink AG1
48:37
48 minutes, 37 seconds
helps me fill in nutritional gaps while supporting energy, digestion, and immune function. It also has prebiotics and
48:44
48 minutes, 44 seconds
probiotics to promote gut health. I take my health very seriously, and so should you. Try AJ1 today by visiting drink
48:52
48 minutes, 52 seconds
ag.com/smr or click the link in the pin comment and enjoy a free welcome kit with vitamin D3
49:00
49 minutes
plus K2 and AG1 travel packs. That’s drinkagg1.com/smr.
Sync to video time