# AI News Summary - December 2, 2025

## OpenAI declares 'code red' as Google catches up in AI race
- **OpenAI's 'code red'** signals a critical shift as CEO Sam Altman prioritizes enhancements to ChatGPT, reflecting the urgency of competition from Google and Anthropic.

## Mistral 3 family of models released
- **Mistral 3** introduces a new generation of AI models, including **Mistral Large 3**, which features a **sparse mixture-of-experts architecture** with **41B active** and **675B total parameters**, enhancing performance across multimodal and multilingual tasks.

## Amazon launches Trainium3
- **Amazon's new Trainium3 chip** boasts a **3 nanometer architecture**, delivering over **4x performance** and **40% energy efficiency** improvements compared to its predecessor, enhancing AI training and inference capabilities.

## Apple Releases Open Weights Video Model
- **STARFlow-V** is the **first** normalizing flow-based **causal video generator**, achieving visual quality comparable to diffusion models while enabling **end-to-end training** and **multi-task support** across various video generation tasks, including text-to-video, image-to-video, and video-to-video.

## Anthropic: AI agents find $4.6M in blockchain smart contract exploits
- **AI agents have demonstrated the ability to exploit smart contracts, collectively uncovering vulnerabilities worth $4.6 million in simulated environments, highlighting the economic risks posed by these technologies.**

## Published paper uses hardcoded seed and collapsed model to report fraudulent results
- A **published paper** misrepresented its findings by using a **hardcoded seed** and **collapsed models**, leading to inflated accuracy claims that do not reflect true model performance.

## Claude 4.5 Opus' Soul Document
- **Claude 4.5 Opus' "Soul Document"** is a 14,000 token overview that was used to train the model's personality during its training run, confirmed by Anthropic's Amanda Askell as a real document integral to Claude's development.

## Comparing AWS Lambda ARM64 vs. x86_64 Performance Across Runtimes in Late 2025
- **AWS Lambda's arm64 architecture outperforms x86_64** in cost and performance, with Rust on arm64 being the most efficient combination, achieving up to **40% lower compute costs** while maintaining superior speed across workloads.

## Arcee Trinity Mini: US-Trained Moe Model
- **Arcee AI's Trinity Mini** is a compact, end-to-end trained MoE model designed in the U.S., featuring open weights and advanced reasoning capabilities, with Trinity Large set to launch in January 2026, utilizing **2048 B300 GPUs** for training.

## Qwen3-VL can scan two-hour videos and pinpoint nearly every detail
- **Qwen3-VL** by Alibaba can accurately analyze **two-hour videos** and identify specific frames with **99.5% accuracy**, showcasing its advanced capabilities in handling extensive data loads.

## Pose-free 3D Gaussian splatting via shape-ray estimation
- **SHARE** introduces a **pose-free** framework for **3D Gaussian splatting**, enabling efficient rendering without the need for precise camera poses, thus addressing common challenges in real-world applications.

## Show HN: RunMat – runtime with auto CPU/GPU routing for dense math
- **RunMat** offers a **high-performance runtime** for MATLAB-style math, automatically optimizing operations for CPU or GPU without requiring kernel code or rewrites, achieving performance that can surpass hand-tuned CUDA in many scenarios.

## Gated Attention, a bit of schmidhubering/sociology of science
- **Gated Attention** has emerged as a significant concept, particularly through its **headwise gating** mechanism, which applies a dense 
[0,1] coefficient to each attention head, enhancing output mixing efficiency.

## At NeurIPS, NVIDIA Advances Open Model Development for Digital and Physical AI
- **NVIDIA introduces the DRIVE Alpamayo-R1**, the first open industry-scale reasoning vision language action model for autonomous driving, enhancing vehicle decision-making in complex scenarios through advanced AI reasoning techniques.

## Polymathic release new scientific foundation model - paper shows it learns general abstract laws of physics
- **Polymathic AI's** new model, **Walrus**, demonstrates an ability to grasp **abstract physical concepts** such as speed and diffusion, indicating a significant leap toward creating general-purpose scientific AI.