# Nov 18, 2025

## Daily

### Gemini 3
- **Gemini 3** is Google's most advanced AI model, enhancing **reasoning** and **multimodal capabilities**, enabling users to bring any idea to life across various Google products like the Gemini app and Vertex AI.

### WeatherNext 2: Our most advanced weather forecasting model
- **WeatherNext 2** is Google's latest AI model, delivering **faster**, **more accurate**, and **higher-resolution** weather forecasts, capable of generating hundreds of scenarios in under a minute using a single TPU.

### Solving a Million-Step LLM Task with Zero Errors
- **MAKER** is the first system to solve a task with **over one million LLM steps** without errors, showcasing a breakthrough in long-range task execution.

### Gemini 3 for developers: New reasoning, agentic capabilities
- **Gemini 3 Pro** is Google's most advanced model, excelling in **AI benchmarks** and enabling **agentic coding** through platforms like Google Antigravity and the Gemini API.

### 5 Things to Try with Gemini 3 Pro in Gemini CLI
- **Gemini 3 Pro** enhances **Gemini CLI** by integrating advanced reasoning and agentic coding, enabling users to generate complex applications and workflows directly from the terminal.

### Google Brings Gemini 3 AI Model to Search and AI Mode
- **Gemini 3** enhances Google Search with **state-of-the-art reasoning** and **multimodal understanding**, enabling dynamic visual layouts and interactive tools tailored to user queries.

### LeJEPA
- **LeJEPA** introduces a **theoretically grounded** training objective for Joint-Embedding Predictive Architectures (JEPAs), utilizing **Sketched Isotropic Gaussian Regularization (SIGReg)** to optimize embedding distributions for improved prediction accuracy.

### Trying out Gemini 3 Pro with audio transcription and a new pelican benchmark
- **Gemini 3 Pro** enhances capabilities from Gemini 2.5, featuring **1 million input tokens** and **64,000 output tokens**, with improved performance in multimodal tasks including audio transcription and image processing.

### Microsoft, NVIDIA and Anthropic Announce Strategic Partnerships
- **Microsoft, NVIDIA, and Anthropic** have formed a strategic partnership to scale Anthropic's Claude AI model on Azure, with Anthropic committing to purchase **$30 billion** in Azure compute capacity and up to **1 gigawatt** of additional capacity.

### [P] DeepClause - A Neurosymbolic AI System
- **DeepClause** is a **neurosymbolic AI system** that integrates **symbolic reasoning** with **neural language models**, enabling agents to perform complex logic and multi-step reasoning through its **Prolog-based DSL**, DML (DeepClause Meta Language).

### [R] Unlocking Out-of-Distribution Generalization in Transformers via Recursive Latent Space Reasoning
- The paper presents **strong results in out-of-distribution (OOD) length generalization** through a combination of variable recurrence, step-wise curriculum training, and discrete bottlenecks, achieving over **4x improvement** on simple math tasks.

### OpenGuardrails: open-source AI safety and guardrail platform released
- **OpenGuardrails** is the first fully open-source platform designed to ensure the **safety, robustness, and privacy compliance** of large language models (LLMs) by addressing content-safety violations, model-manipulation attacks, and data leakage risks.

### NVIDIA Accelerates AI for Over 80 New Science Systems Worldwide
- **NVIDIA has unveiled over 80 new scientific systems globally, achieving a remarkable total of 4,500 exaflops of AI performance**, with the latest addition being the 300-petaflop Horizon supercomputer at TACC, set to revolutionize scientific research by 2026.

### Accelerated Computing, Networking Drive Supercomputing in Age of AI
- **NVIDIA's innovations** at SC25, including the **DGX Spark** and **BlueField-4 DPUs**, showcase a leap in AI supercomputing capabilities, enabling developers to run complex models with unprecedented efficiency and performance.

### NVIDIA Accelerated Computing Enables Scientific Breakthroughs for Materials Discovery
- **NVIDIA's accelerated computing** through platforms like **Holoscan** and **ALCHEMI** is revolutionizing materials discovery, enabling real-time nano-imaging and high-throughput simulations that enhance research efficiency and innovation in industries such as aerospace and energy.