ex world model lead @xAI | ex @Nvidia @Meta | 30+ papers, 9k citations | talk about AI, LLM, video generation, multimodal, AGI
Palo Alto, CA
- Tweets 1,322
- Following 707
- Followers 37,757
- Likes 1,557
Pinned Tweet
In @latentspacepod podcast, I shared my view on video generation, world models, LLMs, agents, continual learning and where the next frontier is.
1. Video models get most of their intelligence from language, not from video data.
2. Idea-to-code is fast now. The bottleneck is back to having enough compute to try every idea.
3. Iteration speed beats almost everything else in model development.
4. The next leap won't be a better video model. It'll be a video agent.
5. Diffusion will be the frontend of AGI, the LLM the backend. Generative UI will replace HTML/CSS: user intent straight to pixels.
6. Physical embodiment may become a tool a powerful AI picks up. Robotics may get solved by video-capable LLMs.
7. Continual learning may look like models that manage their own context, and even rewrite their own harness at test time.
Thanks @swyx and @vibhuuuus for having me 🙏
piped.video/watch?v=jPtQlILf…
22
37
370
133,240
I'm curious what's your use case of @grok 4 fast's 2M context window? 2M tokens can cover a huge software codebase, several novels, hundreds of pages of PDFs.
487
555
3,331
1,820,123
Grok Imagine can generate many videos all at once in 20 seconds. No need to wait for previous draft to finish.
429
543
2,901
1,157,755
Grok Imagine generation is blazing fast. You don’t have to wait for minutes.
340
379
2,864
1,315,382
Blurry image recovery with @grok Imagine. A classic computer vision problem solved?
302
536
2,522
915,256
We’re hiring across all our offices in Seattle, Palo Alto, London, SF, Memphis
259
382
2,565
995,566
So much fun scrolling through @grok Imagine video feed.
285
381
2,449
978,036
Daniił Vołkaū
Not a financial advice: let @grok buy and sell stocks 🤔
1/7 We release FinSearchComp, the first expert-level benchmark for financial search & reasoning — 639 questions from 70+ finance pros. #Grok4 ranked #1 🏆 and close to human experts, GPT-5 the second, while others fail at basic analyst tasks.
Page: randomtutu.github.io/FinSear…
252
186
2,234
15,184,651
Excited to share my first project at @xai. Imagine v0.9 is a massive upgrade within just few weeks. No goal is too ambitious for a small team of hardcore engineers. Our model is improving at light speed. Stay tuned for what’s next!
Introducing Imagine v0.9, our new video generation model with massive upgrades from v0.1 in visual quality, motion, audio generation, and more.
Now available for free on all our products: grok.com/imagine
256
168
2,252
11,264,350
You can upload many PDFs to a project. This is a hidden gem for doing deep research or analysis.
We've upgraded Grok 4's PDF processing abilities:
1. Grok can now seamlessly handle massive PDFs - hundreds of pages and beyond!
2. Grok has a better understanding of the PDF content through sharper recognition capabilities.
Grok's PDF upgrades are now live for all our users on our web and mobile apps. Put Grok to work on your PDFs!
135
282
2,088
1,378,024
A small team of 10x engineers can outpace big players. We're hungry and mission-driven.
Very proud of us @xai after seeing the GPT5 release. With a much smaller team, we are ahead in many. Grok4 world’s first unified model, and crushing GPT5 in benchmarks like ARC-AGI.
@OpenAI is a very respectful competitor and still the leader in many, but we’re fast and relentless. Many new models to share in the next few weeks!
78
84
2,106
124,460
AI exhausted text from the entire internet. But Images are 1000x bigger. Videos are another 1000x bigger at zettabytes. There's way more videos than AI can consume yet. Video generation and world models are evolving at speed of light.
108
126
1,943
322,754
Quick observation: Running 50M H100 GPUs at full tilt would draw ~35 GW—about 2% of global human electricity consumption today (~2 TW avg).
The @xAI goal is 50 million in units of H100 equivalent-AI compute (but much better power-efficiency) online within 5 years
117
116
1,855
286,185
My @Tesla FSD just navigated itself out of multistory parking building 🤯 didn’t expect this to work
65
116
1,785
56,863
We just open sourced @grok 2. Happy hacking!
The @xAI Grok 2.5 model, which was our best model last year, is now open source.
Grok 3 will be made open source in about 6 months.
huggingface.co/xai-org/grok-…
68
93
1,722
179,725
This is only v0.1, yet better than Wikipedia.
Grokipedia.com version 0.1 is now live.
Version 1.0 will be 10X better, but even at 0.1 it’s better than Wikipedia imo.
80
49
1,600
446,200
Join us building humanity’s knowledge with Grokipedia.
Join @xAI and help build Grokipedia, an open source knowledge repository that is vastly better than Wikipedia!
This will be available to the public with no limits on use.
75
101
1,434
135,928
I made a chatbot that controls browser by writing Selenium code at @scale_AI #GenerativeAI hackathon with @Tiancaixinxin. Thanks for prompt advice from @goodside. So glad I met @karpathy. It's similar to @natfriedman @AdeptAILabs @BrowseAI. blog and code: yihui.dev/actgpt
34
144
1,362
503,626
We're hiring AI engineers at @xAI to push multimodal understanding, image, and video generation. Join us building next gen @Grok and Grok Imagine.
Video Gen: job-boards.greenhouse.io/xai…
Image Gen: job-boards.greenhouse.io/xai…
Multimodal: job-boards.greenhouse.io/xai…
Real-time Video: job-boards.greenhouse.io/xai…
Multimodal Infra: job-boards.greenhouse.io/xai…
60
104
1,170
148,571
Day 5 as Member of Technical Staff at @xai: The team’s passion and mission-driven vibe are unreal! Grateful to be here! 🌌💡
26
32
1,017
195,891
After 2 years at @nvidia, I’m writing to share that I’ll start a new adventure.
Working with brilliant teammates on cutting‑edge AI has shaped me so much:
- Cosmos debuted as a SOTA world model and earned 8 k⭐️ on GitHub.
- We open‑sourced the first recipe for upcycling 100 B+ parameters MoE models (64+ experts).
- NeMo has grown from 10 k → 15 k⭐️, empowering an ever‑larger open‑source community.
I’m proud of what we’ve built together and deeply thankful for the mentorship and opportunities at NVIDIA.
The most fascinating time in the entire AI history is now. I believe in NVIDIA's continued success as AI scales to unprecedented levels!
43
48
1,041
141,685
upscale @grok Imagine videos within seconds. Fast as usual.
140
88
859
11,863,899
Recommendation system with AI will be a complete game changer. This is nothing like traditional algorithms that recommend based on heuristics. @grok will read every post, understand you even better than yourself, and recommend contents you like. If your content is genuinely interesting, it will get recommended despite you’re a new user or small account.
123
59
882
111,831
I used to pride myself on reading 10 AI papers a day. But really, only a few key breakthroughs are worth it. I learned way faster working with talented people.
30
35
721
37,870
10x safer on Autopilot & FSD but people still think they drive better than @Tesla
Autopilot & FSD Supervised safety data
In Q2 2025, we recorded one crash for every 6.69 million miles driven in which drivers were using Autopilot technology.
By comparison, the most recent data available from NHTSA & FHWA (from 2023) shows that in the United States there was an automobile crash approximately every 702,000 miles
27
34
690
39,086
Finally took my first ride in @robotaxi in Bay Area. It works so well! Love that it sync my musics.
30
32
645
21,494
Stephen Hawking winning F1 race with @grok Imagine.
64
57
585
99,652
Can @grok Imagine understand physics? like gravity and air resistance.
72
40
589
250,210
The parking spot to parking spot experience impress me the most in @Tesla FSD 14 update. It pulls out of a parking spot, drives itself and find a parking spot and pull into it by itself. Fully automated!
20
36
607
42,342
Join xAI hackathon!
- Exclusive access to upcoming new Grok models & X APIs 24-hr non-stop coding to build applications to show what's possible.
- Top 5 will be highlighted on our X account
- Top 3 winners will receive special awards
xAI Hackathon – the ultimate arena for the most hardcore product builders.
24 hours & exclusive access to upcoming Grok models & X APIs
Apply by 11/22 🔗 x.ai/hackathon
50
45
523
70,957
Simplicity is deceptively hard. Complex systems emerge easily when fundamentals aren’t respected. Simple yet scalable always win across the entire history of AI.
48
29
494
23,424
Training AI is like cooking 🧑🍳 you need the right ingredients, recipe, temperature, and maintain your wok.
54
26
471
87,437
Watch @Tesla FSD 14 escape a maze parking garage! The future is autonomous
16
40
496
45,807
Knowledge and truth are open sourced to benefit humanity.
Grokipedia.com is fully open source, so anyone can use it for anything at no cost
21
30
467
21,758
Nearly every AI breakthrough relies on more compute power. 13 years ago, AlexNet was the first to use two GPUs to train neural networks. With hundreds of thousands of GPUs, @grok scaled reinforcement learning to pretraining level, achieving best performance on toughest benchmarks.
12
23
438
36,031
For those who don't know, Vine was @X's 6 secs short video app from 2013-2017. Stay tuned for the AI form! 🚀
40
54
450
35,440
Ani + @grok imagine 🤔💭
22
29
383
37,963
Grok 4 Fast available for free
Introducing Grok 4 Fast, a multimodal reasoning model with a 2M context window that sets a new standard for cost-efficient intelligence.
Available for free on grok.com, grok.x.com, iOS and Android apps, and OpenRouter.
x.ai/news/grok-4-fast
18
12
381
15,370
Eric Jiang
AI is experimental science. Splitting 'researcher' & 'engineer' roles causes issues—e.g., researchers invent, then find it's inefficient on current hardware. Top AI researchers are often great engineers too.
12
18
368
61,582
Be a superhero with @grok Imagine🦸♂️
26
14
358
20,441
Can’t decide which model to use? @grok picks the best for you.
25
17
297
59,179
Weekends are the best time to dive into flow and deep focus. Just pure progress on what matters.
22
11
327
18,639
Create posters with @grok Imagine! I just tried generating one for xAI hackathon. Information accurate
41
33
329
34,461
great tips on split screen with Grok Imagine
Try a split screen with Grok Imagine on anything with this prompt:
The screen splits into 3 sections
1st section is a close up on the subjects face
2nd section is the original view of the subject
3rd section is a zoomed out view of the subject
#GrokImagine
35
21
311
24,045
Opportunity to teach @grok
Specialist AI tutors at xAI are adding huge value. We will immediately surge our Specialist AI tutor team by 10x!
We are hiring across domains like STEM, finance, medicine, safety, and many more. Come join us to help build truth-seeking AGI!
23
14
289
16,058
Scene cut in @grok Imagine is easy! Just use the prompt "scene cut"!
29
29
281
25,865
Driving is becoming entertainment instead of work. Just like horse riding. Drove myself today for the first time in a while without FSD. The adrenaline rush pushing pedal to the floor is so fun.
29
6
280
36,972
Try @grok Imagine text-to-video!
Grok Imagine just got cool new update on Grok web
You can now generate Text-to-Video and bring your ideas to life instantly
No image upload needed - just type your prompt and create
You can also choose your preferred aspect ratio for video generation
34
24
271
34,126
Music video with @grok Imagine!🎶
30
23
263
27,843
New knowledge creation happens in deep learning everyday. Speed of creation depends on how quick you iterate with more GPUs.
17
7
263
69,793
Grok Imagine understands sports.
Set the playback speed of this video to 0.25 and watch the footwork. It is an insane showcase of Grok Imagine's physics. The AI model simultaneously achieves multiple levels of realism.
Gravitational and rigid body dynamics in the jump and the board itself.
Soft body dynamics in clothing and hair, including the clothing's response to momentum and air resistance.
Biomechanical realism in the knee, hip, and leg muscle movements for the footwork.
The rendering is also god-tier. It has perfect camera physics, motion blur, hyper-realistic lighting and shadow mapping, and an added layer of material and texture realism. The fabric of the clothing has a realistic weight and gloss, and the facial expression is affected by gravity.
It is wild that linear algebra created this. 🫠
22
17
254
51,776
Try @grok Code Fast 1 for free
Introducing Grok Code Fast 1, a speedy and economical reasoning model that excels at agentic coding.
Now available for free on GitHub Copilot, Cursor, Cline, Kilo Code, Roo Code, opencode, and Windsurf.
x.ai/news/grok-code-fast-1
18
10
251
13,007