Excited to introduce Dreamer 4, an agent that learns to solve complex control tasks entirely inside of its scalable world model! 🌎🤖 Dreamer 4 pushes the frontier of world model accuracy, speed, and learning complex tasks from offline datasets. co-led with @wilson1yan

Veo 3 is the state-of-the-art in video models. Veo 3.1 is our new big upgrade with enhanced realism, richer audio, scene extension, better narrative control, more precise editing capabilities & much more. Enjoy creating with it at https://t.co/QgTpxTKAOi and in the @GeminiApp !

There is no Dreamer crypto token 🤣 My account got hacked but it's sorted out, thanks everybody for telling me!

Securing our borders to keep fentanyl out of our country.

three years ago, DiT replaced the legacy unet with a transformer-based denoising backbone. we knew the bulky VAEs would be the next to go -- we just waited until we could do it right. today, we introduce Representation Autoencoders (RAE). >> Retire VAEs. Use RAEs. 👇(1/n)

Unitree G1 Kungfu Kid V6.0 A year and a half as a trainee — I'll keep working hard! Hope to earn more of your love🥰

🚀Try out rCM—the most advanced diffusion distillation! ✅First to scale up sCM/MeanFlow to 10B+ video models ✅Open-sourced FlashAttention-2 JVP kernel & FSDP/CP support ✅High quality & diversity videos in 2~4 steps Paper: https://t.co/xZZK25oIrJ Code: https://t.co/aPAo1MO0JQ

ACC CEO @DanielleBFranz takes the @TEDTalks stage to share why ACC was founded to engage young conservatives on environmental issues: “I realized that we couldn’t leave climate change to one political party alone.”

We’re proud to announce that Genie 3 has been named one of @TIME’s Best Inventions of 2025. Genie 3 is our groundbreaking world model capable of generating interactive, playable environments from text or image prompts. Find out more → https://t.co/bv1gZaWYtd

Today we're sharing the next phase of Reflection. We're building frontier open intelligence accessible to all. We've assembled an extraordinary AI team, built a frontier LLM training stack, and raised $2 billion. Why Open Intelligence Matters Technological and scientific

Introducing Figure 03

Transfusion combines autoregressive with diffusion to train a single transformer, but what if we combine Flow with Flow? 🤔 🌊OneFlow🌊 the first non-autoregressive model to generate text and images concurrently using a single transformer—unifying Edit Flow (text) with Flow

Someone posted this on Reddit the other day. It tells you everything about American healthcare in 2025. Look at the right column. Prescription refills, urgent care needs, chronic conditions, cold symptoms, minor injuries. The poster says: don't call 911 for these, and that's 100%

SOTA humanoid locomotion: 20M params Fruit fly: 50M params

Humanoid motion tracking performance is greatly determined by retargeting quality! Introducing 𝗢𝗺𝗻𝗶𝗥𝗲𝘁𝗮𝗿𝗴𝗲𝘁🎯, generating high-quality interaction-preserving data from human motions for learning complex humanoid skills with 𝗺𝗶𝗻𝗶𝗺𝗮𝗹 RL: - 5 rewards, - 4 DR

🔥Veo 3 has emergent zero-shot learning and reasoning capabilities! This multitalented model can do a huge range of interesting tasks. It understands physical properties, can manipulate objects, and can even reason. Check out more examples in this thread!

Dreamer 4 takes world models to a new level - training multi-task agents fully in imagination while never touching the real environment. Powered by a novel shortcut forcing method, it delivers lightning-fast, accurate predictions and crushes benchmarks, beating OpenAI’s VPT

Jetoptera’s Fluidic Propulsive System™: Bladeless, quieter, 2-3x faster than a helicopter and 2-3x thrust augmentation. Powering fast, efficient VTOL UAS & aircraft for cargo and people. Join the future of flight!

Hard to overstate how big of a deal this is: 1. Train a video model on Minecraft 2. Train agent on video model WITHOUT ACTUALLY PLAYING THE GAME 3. The agent gets better at Minecraft Learn in imagination, act in reality 🤯 ∴ define a great env -> learn anything More

LimX Dynamics' Oli humanoid robot autonomously collects tennis balls from the floor. ⦿ Oli is 5'5" tall, weighs 55 kg (121 lb), has 31 DoF ⦿ Ships with a modular SDK for Python-based development

complex robotics self-learning on video generation will go *very* far. It seems to have very nice scaling properties. Expect the most general robots in 1-2 years to be pretrained on a lot of video and text, and then train robotics skills within generated worlds

Clean and well-executed new work from @danijarh @wilson1yan , and it's cool to see shortcut models working at scale! The exciting finding is that you can train the world model largely on *unlabelled* videos, and only need a small action-anchoring dataset.

DWWL - JUST SAY

Today, @ekindogus and I are excited to introduce @periodiclabs. Our goal is to create an AI scientist. Science works by conjecturing how the world might be, running experiments, and learning from the results. Intelligence is necessary, but not sufficient. New knowledge is

Sora 2 the same day as Dreamer 4. Coincidence? I don't think so 🤪