Google DeepMind

We’re welcoming a new 𝟮 𝗯𝗶𝗹𝗹𝗶𝗼𝗻 𝗽𝗮𝗿𝗮𝗺𝗲𝘁𝗲𝗿 𝗺𝗼𝗱𝗲𝗹 to the Gemma 2 family. 🛠️ It offers best-in-class performance for its size and can run efficiently on a wide range of hardware. We’re also introducing 𝗦𝗵𝗶𝗲𝗹𝗱𝗚𝗲𝗺𝗺𝗮: a series of state-of-the-art safety classifiers designed to filter harmful content. 🛡️ These target hate speech, harassment, sexually explicit material and more, both in the input and output stages. Finally, we’re announcing 𝗚𝗲𝗺𝗺𝗮 𝗦𝗰𝗼𝗽𝗲, a set of tools to help researchers examine how Gemma 2 makes decisions. 🔍 Developers can get started with 2B today → https://dpmd.ai/4d0MKEH

You can now access 1.5 Flash in Gemini. ⚡ Optimized for speed and efficiency, it comes with a 32K context window - meaning you can ask more complex questions while getting faster answers. 💬 Try it in over 40 languages and 230 countries and territories. → https://dpmd.ai/3YjADhy

We’re presenting the first AI to solve International Mathematical Olympiad problems at a silver medalist level.🥈 Here’s why this is important. ↓ 🧮 The IMO is the world’s most prestigious annual math competition, presenting six problems to participants involving: geometry, algebra, number theory, and combinatorics. Our hybrid system consists of two breakthrough models: ➕AlphaProof, which can generate complex proofs using Lean, a math programming language. This allows answers to be formally verified for correctness. ➕AlphaGeometry 2: an improved version of our previous system to tackle geometry. Trained on more synthetic data than its predecessor, it managed to solve this year’s IMO Problem 4 within 19 seconds. We’re excited to see how our new system could help accelerate AI-powered mathematics, from quickly completing elements of proofs to eventually discovering new knowledge for us - and unlocking further progress towards AGI. Find out more → https://lnkd.in/d6TUU8_Q

Hello from [ICML] Int'l Conference on Machine Learning in Vienna! 🇦🇹 At our booth, we’re hosting live research demos on Gemini Nano, LearnLM - our new family of AI models for education - and TacticAI, an AI assistant that can help with football tactics. We’re also covering themes like exploring AGI, the challenges of scaling and the future of multimodal generative AI in our oral, spotlight and poster presentations. We’re proud to sponsor ICML and help create a more diverse community in AI by supporting initiatives led by Disability in AI, Queer In AI, Women in Machine Learning and Latinx in AI (LXAI). Come say hi to us and our Google Research colleagues next door! 👋 See our schedule → https://dpmd.ai/4cQ8Cma

Heading to [ICML] Int'l Conference on Machine Learning in Vienna next week? 🇦🇹 We’ll be presenting over 8️⃣0️⃣ papers and hosting live demos on AI assistants for coding, football tactics, and more. You can also find out more about Gemini Nano, our most efficient model for on-device tasks - as well as LearnLM, our new family of models for education. → https://lnkd.in/e27fuegt

Our latest research uses Gemini 1.5 Pro's model’s long context window to help our robots find their way around the world. 🤖 “Gemini’s ability to handle video and text… allows the “Google helper” robot to make sense of its environment and navigate correctly when given commands that require some commonsense reasoning.“ This work could represent the next step in human-robot interaction. 🤝 Find out more in WIRED → https://dpmd.ai/3WnBtIv

Snowy scenes created by Veo. ❄️✨ What could be next in this story? 👀

Quanta wrote a really wonderful article on the history of protein structure prediction, and they captured great stories and quotes from scientists all around the field. It even includes a picture of me from my undergrad days looking like a "real" physicist. I would highly recommend it!

“Hey robot, take me somewhere I can draw?” 🤖 We challenged our helper robots to navigate their way around a busy space - using Gemini 1.5 Pro. With the model’s 1 million token context window, it’s able to recall an environment after watching a video tour, and successfully followed a range of instructions - from finding a specific desk to remembering a favorite drink. Find out more in our latest paper → https://dpmd.ai/4bUobbj

We’re thrilled to have won the prestigious MacRobert award from the Royal Academy of Engineering for our work on GraphCast - our AI model delivering 10-day weather forecasts in under one minute, with unprecedented accuracy via deep learning and vast data sets. GraphCast has revolutionized AI-powered weather forecasting, significantly enhancing the speed and accuracy of predictions. This new approach has the potential to benefit billions worldwide, empowering critical decision-making across industries, optimizing resource allocation, and mitigating the impact of severe weather events. By enabling authorities to issue timely safety and evacuation warnings, AI-powered predictions could help save lives and protect communities. Read more about the award ↓ https://lnkd.in/ecKddvhD