Google DeepMind

A paper origami owl, a felt robot, a skater made of clay… here’s how Imagen 3 can generate visually rich images with complex textures. 👀 Join the waitlist to access it through ImageFX - part of a new suite of AI tools in Labs → https://dpmd.ai/3K9GGg4

Turn your idea into a new alphabet with GenType and Imagen 2. 🔤 🎨 With one simple prompt, you can create a whole set of creative, customized letters. Try it today → labs.google/gentype

Turn your idea into a new alphabet with GenType and Imagen 2. 🔤 🎨 With one simple prompt, you can create a whole set of creative, customized letters. Try it today → labs.google/gentype

I was honored to speak at AI for Good Global Sunmit last Thursday about how AI can help us scale our efforts to monitor, mitigate and adapt to climate change. I shared a few concrete examples where AI already enhances human capabilities and accelerates complex modelling of decades-old problems. For example, AI is already changing weather modelling. Google DeepMind ’s model GraphCast can forecast weather of up to 10 days with unprecedented accuracy and speed, which can help predicting extreme weather to help us prepare earlier. Similarly, our OPF AI model predicts optimal power flows hundreds to thousands of times faster than traditional methods, potentially helping to simulate multiple demand peaks and supply lows that are inherent in new energy systems. With improved speed and precision we can advance our ability to make better decisions. Leveraging AI for climate action will require data that is better available and usable, and people who have both AI and climate science capabilities. Responsible collaboration across the ecosystem is needed to make this happen, with a consistent deployment of safety and responsibility practices. Thank you to International Telecommunication Union for inviting me to speak. You can watch my session linked in the comment below.

The Gemini app is rolling out to the UK and countries across Europe - putting the power of AI into more people's hands. 🌍 ✍️ Get help writing thank you notes 🍜 Find inspiration for your next meal 🛞 Snap a photo of a flat tire - and ask how to change it And more. → https://dpmd.ai/4bOfsbe

Our highest quality text-to-image model Imagen 3 can create more photorealistic visuals - including hands, fingers, faces and groups of people. 👐 Join the waitlist to access it through ImageFX - part of a new suite of AI tools in Labs at Google → https://dpmd.ai/3K9GGg4

“We have entered an age where a single human mind cannot comprehend the data that we are gathering about the universe.” 🌌 On the latest episode of Andreessen Horowitz a16z podcast, our VP of Research for Science Pushmeet Kohli talks about how AI can help scientists, mathematicians and others discover new insights, process huge amounts of data and make breakthroughs in their fields. Listen now ↓ https://lnkd.in/daqntx4x

What kind of videos can Veo generate from just a single image? 👀 Our model can create clips that follow the original visual’s style as well as instructions from a text prompt. Find out more → https://dpmd.ai/li-veo

The final weeks of the football season are always nerve-wracking, so it's a nice change of pace to see the beautiful game inspiring innovation off the pitch. One of our recent projects explored how robots can emulate the dynamic skills we see in football players - turning, kicking, and the strategic pursuit of the ball. Leveraging our MuJoCo physics engine, we trained AI agents in basic football skills, progressing from simulations to real-world application with bipedal robots, each equipped with 20 actuated joints. To train them, we introduced targeted randomness and slight disturbances within the simulator to enable our agents to adapt to unpredictable real-world dynamics. The result? Our robots showcased agility and dynamism that surpassed traditional programming, mastering complex manoeuvres from goal-scoring to defending, as well as demonstrating a fundamental grasp of one-on-one gameplay. While a robot kickabout is good fun, this research is an important step toward training general robots rather than training them for specific tasks, helping bring us closer to our goal of safely integrating robots into our daily lives.   Check out the video below: https://lnkd.in/gqEp76hk

“We need a technology that will harness AI so that everyone can use it and no one is left behind.” In India, developers used Gemma to create Navasara: a model that’s trained for 15 Indic languages. 🇮🇳 ↓