World Wire

Most AI systems today work in turns. You type or speak, the model waits, processes your input, and then responds. That’s the entire interaction loop. Thinking Machines Lab, an AI research lab, is arguing that this model of interaction is a fundamental bottleneck. Thinking Machines Lab team introduced a research preview of a new class of system they call interaction models to address it. The main idea for their research is interactivity should be native to the model itself, not bolted on as an afterthought. What’s Wrong with Turn-Based AI If you’ve built anything with a language model or voice API, you’ve worked around the limitations of turn-based interaction. The model has no awareness of what’s happening while you’re still typing or speaking. It can’t see you pause mid-sentence, notice your camera feed, or react to something visual in real time. While the model is generating, it’s equally blind — perception freezes until it fin...

The mouse pointer has sat at the center of personal computing for more than half a century. It tracks cursor position. It registers clicks. Beyond that, it does almost nothing. Google DeepMind researchers outlined a set of experimental principles and demos for an AI-enabled pointer that goes considerably further: one that understands not just where you are pointing, but what you are pointing at and why it matters. The system is powered by Gemini and is currently in the experimental stage. Two demos are live in Google AI Studio today: one for editing an image and one for finding places on a map, both operable by pointing and speaking. A deeper integration called Magic Pointer is also rolling out inside Chrome, and a further integration is planned for Googlebook, Google’s new line of Gemini-powered laptops announced this week. https://ift.tt/rMVCWo2 What DeepMind is Targeting The frustration DeepMind researchers are addressing is a familiar one for anyone who has tr...

A team researchers from China have released AntAngelMed, a large open-source medical language model that the team describes as the largest and most capable of its kind currently available. What Is AntAngelMed? AntAngelMed is a medical-domain language model with 103 billion total parameters, but it does not activate all of those parameters during inference. Instead, it uses a Mixture-of-Experts (MoE) architecture with a 1/32 activation ratio, meaning only 6.1 billion parameters are active at any given time when processing a query. It helps to know how MoE architectures work. In a standard dense model, every parameter participates in processing every token. In an MoE model, the network is divided into many ‘expert’ sub-networks, and a routing mechanism selects only a small subset of them to handle each input. This allows you to have a very large total parameter count — which typically correlates with strong knowledge capacity — while keeping the actual compute cos...

Researchers at Tilde Research have released Aurora , a new optimizer for training neural networks that addresses a structural flaw in the widely-used Muon optimizer. The flaw quietly kills off a significant fraction of MLP neurons during training and keeps them permanently dead. Aurora comes with a 1.1B parameter pretraining experiment, a new state-of-the-art result on the modded-nanoGPT speedrun benchmark, and open codes. What is Muon? To understand Aurora, it helps to first understand Muon. The Muon optimizer attracted attention in the ML community after outperforming AdamW in wall-clock time to convergence on the nanoGPT speedrun competition — a community benchmark that measures how fast you can train a GPT-style model to a target validation loss. Since then, Muon has been adopted in frontier-scale model training by several research groups. Muon’s key algorithmic step is computing the polar factor of the gradient matrix. For a gradient matrix G with thin Singul...

In this tutorial, we explore skfolio , a scikit-learn compatible portfolio optimization library that helps us build, compare, and evaluate different investment strategies in a structured Python workflow. We start by loading S&P 500 price data, converting it into returns, and creating a time-based train-test split suitable for financial analysis. From there, we build simple baseline portfolios, test mean-variance optimization, compare alternative risk measures, apply risk-parity methods, and use hierarchical clustering techniques such as HRP and Nested Clusters Optimization. We also move into more advanced portfolio construction ideas, including robust covariance estimators, Black-Litterman views, factor models, pre-selection pipelines, walk-forward validation, and hyperparameter tuning with GridSearchCV. Copy Code Copied Use a different Browser import subprocess, sys def _pip_install(pkg): subprocess.check_call([sys.executable, "-m", "pip", "install...