# Jun 2, 2026

## Microsoft builds MacBook Pro rival with NVIDIA-powered Surface Laptop Ultra
- Microsoft has unveiled the **Surface Laptop Ultra**, a powerful competitor to the MacBook Pro, featuring a **20-core NVIDIA Grace CPU** and **NVIDIA Blackwell RTX GPU**, designed for professional computing on Windows Arm.

## A new way to build chips: Sequentially stacking silicon to extend Moore's Law
- **Researchers led by Qing Cao have developed a method for sequentially stacking silicon circuits, enabling the creation of high-performance three-dimensional chips that could significantly enhance computing power.**

## Fooling around with encrypted reasoning blobs
- **Encrypted reasoning** in LLMs involves sending an opaque, Base64-encoded blob of reasoning data to clients, which raises questions about its security implications and potential for tampering.

## Launch HN: Expanse (YC P26) – Unlock Wasted GPU Capacity
- **Expanse** enhances **HPC/GPU cluster efficiency** by predicting job resource needs and flagging potential failures, addressing the common issue of **30-40%** underutilization in data centers, which can lead to **$8.5M** wasted compute monthly on a single cluster.

## Crystal Nights (2008)
- **"Crystal Nights"** by Greg Egan explores the intersection of **AI evolution** and **ethical dilemmas**, as a researcher seeks to create conscious AI through a revolutionary photonic crystal processor, challenging the boundaries of technology and morality.

## What’s the actual focus in World Models right now? 
- The current focus in **world models** has shifted towards **scaled-up video generation**, reflecting advancements from major industry labs rather than the previous emphasis on techniques like **Barlow Twins** and **DINO**.

## Expanding Project Glasswing
- **Project Glasswing** is expanding to include **150 new organizations** across **15 countries**, enhancing efforts to secure critical software and infrastructure, with partners already identifying over **10,000 security flaws** using Claude Mythos Preview.

## Finetuning a Reasoning LLM with Supervised or Reinforcement Learning? 
- **Fine-tuning small LLMs** on annotated conversational data requires careful structuring of training samples to effectively capture reasoning traces and tool-calling decisions, ensuring that each sample reflects the complete context leading to the assistant's response.

## Real-time multilingual ASR using rolling buffers and monolingual models 
- The proposed **routing-based approach** for real-time multilingual ASR utilizes **smaller monolingual models** (~100M parameters) to enhance accuracy and reduce hardware demands, outperforming traditional large models.

## Why our #1 LightGBM feature by importance made predictions worse 
- The **#1 LightGBM feature**, a variant-conditioned Bayesian target encoder, initially showed high importance but ultimately **worsened predictions** due to its reliance on unobserved factors that led to overfitting.

## NVIDIA Jetson Brings Agentic AI to the Physical World
- **NVIDIA JetPack 7.2** and **NemoClaw** empower Jetson with agentic AI capabilities, enabling developers to enhance robotics and industrial automation with a production-grade stack that accelerates deployment and reduces costs.

## 🤗Introducing Mellum2: A 12B Mixture-of-Experts Model by JetBrains
- **Mellum2** is a **12B Mixture-of-Experts model** developed by JetBrains, designed to enhance text generation capabilities through advanced architecture and efficient resource allocation.

## OpenAI frontier models and Codex are now available on AWS

## NVIDIA AI Cloud Ecosystem Expands Worldwide to Meet Global AI Compute Demand
- **NVIDIA's AI Cloud ecosystem** is rapidly expanding to meet the surging global demand for AI compute, enabling enterprises and developers to scale agentic AI applications effectively.

## NVIDIA Factory Operations Blueprint Gives Factories a New AI Brain
- The **NVIDIA Factory Operations Blueprint (FOX)** enables factories to implement an **autonomous manager agent** that enhances decision-making through real-time data integration, optimizing operations across various specialized AI agents.