Katya is a wonderful member of our team and I'm humbled to work with her. Thank you for your dedication and passion and here's to many more Googleversaries!

Love this course on building LLMs from Stanford University. So concise and well presented 👏 youtu.be/9vM4p9NN0Ts?si…

Received word about a data leak from Tibber Deutschland containing lots of personal data. Their response? We're sorry. That's really not enough and not appropriate. Rarely seen communication this unfitting for a leak that consists of 50k datasets.

Just discovered httpstat.us which returns a HTTP response you request. You can request by number, like: httpstat.us/404 Randomly from a range and/or list: httpstat.us/random/400-410… And you can set a timeout in ms: httpstat.us/200?sleep=3000

Impressive results for Gemini 🤖

Always wonderful to spend time with the one and only PreusslerBerlin 🇺🇦 -- good to see you, Danny! #DevFest2024

Welcome PaliGemma 2! 🤗 Google released PaliGemma 2, best vision language model family that comes in various sizes: 3B, 10B, 28B, based on Gemma 2 and SigLIP, comes with transformers support day-0 🎁 Saying this model is amazing would be an understatement, keep reading ✨

Wow. Genie 2 is mind-blowing 🤖

So Google DeepMind Gemini 2.0 now > Has 2M token context > Native multimodal input & output > Input modalities: Text, Audio, Image, Video, Document > Output modalities: Text, Audio, Image > Native tools: search, code interpreter > Structured Outputs and JSON function calling >

So great to see all the love for Gemini ❤️

👋 Hey builders - Google AI Studio's Gemini 2.0 is available for you to try out today with: ✅ Cursor ✅ Cody AI ✅ @GitHub Copilot ✅ @e2b_dev ✅ BoltAI ✅ PyCharm, a JetBrains IDE What other extensions, IDEs, and developer tools would you like to see supported?

So much written on DeepSeek. For a great overview, check out Ben Thompson's FAQ. I am looking forward to increased competition; forced constraints drive innovation, and more efficiency makes LLMs economical for more usecases, which will dramatically increase demand.

We are going to witness a “Jevons Paradox” in AI. 🤔 As models get more efficient and use less compute per task, you'd think GPU demand would drop. Plot twist: This just enables bigger models, more access, and more applications, making total GPU consumption skyrocket. The more

Two updates today to Vectara HHEM Leaderboard: 1. Google AI Gemini-2.0-Flash-Thinking-Exp with 1.8% hallucination rate. Impressive given DeepSeek's R1 which has a much higher hallucination rate. 2. IBM Hybrid Cloud & Infrastructure IBM Granite-3.1-8B-Instruct with 8.6%

🥁Introducing Gemini 2.5, our most intelligent model with impressive capabilities in advanced reasoning and coding. Now integrating thinking capabilities, 2.5 Pro Experimental is our most performant Gemini model yet. It’s #1 on lmarena.ai (formerly lmsys.org) leaderboard. 🥇

Gemini 2.5 Pro and Gemma 3 ⚡⚡⚡

To understand a new codebase, my workflow: 1. Convert the repo into markdown: repo2txt.simplebasedomain.com 2. Feed that markdown into Google's AI Studio: aistudio.google.com/prompts/new_ch… Completely free and 200% faster than letting Claude run amok eating tokens. Plus Gemini has a 1Mio context

🚀 Announcing Gemini CLI - your open-source AI agent for the terminal. Powered by Gemini 2.5 Pro. What's inside: 🧬 1M token context 🚦 60 req/min, 1K/day 🔐 Secure shell access 🧭 Web-grounded reasoning 🧰 Plugins + MCP 🗂️ Filesystem I/O 🧑‍🔧 Dev loop ready 📦 Open source -

Gemini CLI is our new #OpenSource AI agent that brings the power of Gemini directly into your terminal! Access Gemini 2.5 Pro with 1M token context window, 60 requests/min, and 1,000 requests/day—at no cost with a free Gemini Code Assist license → goo.gle/3HW0jL0

Gemini CLI is here! Our most powerful open-source CLI that brings Google's Gemini 2.5 models directly into your terminal! With unique features like hierarchical memory (context), self-correcting file edits, and secure sandboxed tool execution. 💡 Hierarchical Memory and