The gold standard of business news

Morning Brew is transforming the way working professionals consume business news.

They skip the jargon and lengthy stories, and instead serve up the news impacting your life and career with a hint of wit and humor. This way, you’ll actually enjoy reading the news—and the information sticks.

Best part? Morning Brew’s newsletter is completely free. Sign up in just 10 seconds and if you realize that you prefer long, dense, and boring business news—you can always go back to it.

Join 4.3 Million Readers Now

Today:

• Microsoft Launches New AI Tools

• DeepSeek AI Model Beats OpenAI

• DeepMind’s AlphaQubit Tackles Quantum Computing Challenges

• US Proposes Manhattan-Style AI Initiative

• Niantic Builds AI with Pokémon Go Player Data

Satya Nadella "THE NEW SCALING LAWS" and the Agentic Future [Microsoft Ignite 2024 SUPERCUT]

Microsoft is enhancing productivity with its AI-powered co-pilot tools integrated into apps like Word, Excel, and Teams. These AI tools act as personal assistants, automating tasks, improving creativity, and saving time. Unlike Moore’s Law, which predicted computer growth, AI performance is improving even faster. 

The new features include handling different media types, better problem-solving, and long-term memory. Businesses can easily create custom AI helpers using simple tools, while Microsoft ensures they are secure and effective. 

WATCH THE VIDEO ON YOUTUBE

DeepSeek AI Model Beats OpenAI

DeepSeek, a branch of the Chinese hedge fund High-Flyer Capital Management, has launched its new AI model, R1-Lite-Preview. This advanced model, available through DeepSeek Chat, excels in reasoning tasks like math and logic, sometimes performing better than OpenAI’s latest models. R1-Lite-Preview shows its thinking process step-by-step, helping users understand how it reaches answers. It successfully answers tricky questions that other AI struggled with. 

While the model is free to use with some limits, DeepSeek plans to make it open-source soon. This release continues DeepSeek’s commitment to creating accessible and transparent artificial intelligence technologies.

DeepMind’s AlphaQubit Tackles Quantum Computing Challenges

Google DeepMind and Quantum AI introduced AlphaQubit, an AI-based decoder designed to identify and correct quantum computing errors with remarkable accuracy. Quantum computers, capable of solving complex problems far faster than classical ones, often struggle with errors caused by noise. 

AlphaQubit uses neural networks to analyze error patterns, outperforming traditional methods in accuracy and scalability. Tested on Google’s Sycamore quantum processor, it showed superior performance across various tasks and simulated systems. While promising, AlphaQubit remains too slow for real-time correction on large-scale quantum systems. 

US Proposes Manhattan-Style AI Initiative

A U.S. congressional commission has proposed a large-scale initiative similar to the World War II Manhattan Project to develop advanced artificial intelligence (AI) systems that match or surpass human intelligence. This move comes amid rising competition with China over cutting-edge technologies. The commission emphasizes the importance of collaboration between the government and private companies to advance AI, though it did not specify exact funding plans. 

Additionally, it recommends removing trade exemptions for low-value Chinese goods to tighten security. The initiative aims to ensure the U.S. leads in AI development, enhancing national security and technological superiority.

Niantic Builds AI with Pokémon Go Player Data

Niantic is using data from its Pokémon Go game and the Scaniverse app to develop an AI navigation system called the Large Geospatial Model (LGM). This model processes physical spaces using over 10 million geolocated scans, collected primarily by players. By training more than 50 million neural networks with 150 trillion parameters, Niantic aims to map global locations with remarkable detail and accuracy, even from unfamiliar angles. 

The technology supports augmented reality, robotics, and spatial planning. While some users are unsurprised their data is used this way, it highlights the growing role of AI in geospatial and AR technologies.

🧠RESEARCH

RedPajama: an Open Dataset for Training Large Language Models

RedPajama, an open dataset addressing challenges in training transparent and effective large language models. It includes over 100 trillion tokens, quality signals, and metadata, enabling better data filtering. RedPajama supports creating strong models like Snowflake Arctic and Salesforce's XGen, promoting transparency and innovation in AI development.

Continuous Speculative Decoding for Autoregressive Image Generation

Continuous Speculative Decoding (CSpD) to speed up continuous-valued autoregressive image generation models. By adapting speculative decoding to continuous spaces and addressing sampling challenges, the method achieves a 2.33x speed boost while preserving output quality. It offers practical improvements for efficient and accurate image generation.

Soft Robotic Dynamic In-Hand Pen Spinning

SWIFT, a soft robotic hand system for dynamic in-hand tasks like pen spinning. Using real-world trial-and-error without prior object knowledge, SWIFT achieves 100% success on various pens and generalizes to other objects, showcasing potential for versatile dynamic manipulation.

StableV2V: Stablizing Shape Consistency in Video-to-Video Editing

StableV2V, a video editing method ensuring shape consistency by aligning motion patterns with user prompts. It edits frames sequentially and propagates changes, achieving superior visual consistency and efficiency. A new benchmark, DAVIS-Edit, validates its performance across varied prompts and challenges.

VeGaS: Video Gaussian Splatting

VeGaS, a video editing model using Folded-Gaussian distributions to capture nonlinear video dynamics. It enables realistic video modifications and outperforms current methods in frame reconstruction. VeGaS expands editing possibilities beyond basic transformations.

🛠️TOP TOOLS

HeyGen IOS - AI-powered tool that transforms your videos, voice, and text into lifelike avatars

Grantable - AI designed for grants. 

Vozo Video Translator - Precise Video Translation, Perfected with Prompts

Listening - Listen to academic papers on the go

Leap AI - Build custom AI automations to scale your marketing, sales, and operations—making your team 10x more efficient, without coding.

📲SOCIAL MEDIA

🗞️MORE NEWS

• Meta has updated Messenger with HD video calls, AI noise suppression, Siri integration, and customizable AI video call backgrounds. These features enhance communication, following last year’s end-to-end encryption update.

• OpenAI and Common Sense Media launched a free one-hour course for K-12 educators, "ChatGPT Foundations," teaching AI basics, ethical use, and classroom integration to improve teaching and learning outcomes effectively.

• KPMG will invest $100 million in a Google Cloud partnership to enhance AI services, focusing on AI agent development and workforce training, highlighting professional services' shift towards large-scale AI adoption.

• Israeli startup aiOla launched Whisper-NER, an open-source AI model combining transcription and privacy-focused data masking. It identifies and hides sensitive information like names in real-time, enhancing compliance and security in audio transcription workflows.

• Anthropic's new paper outlines statistical methods to improve AI model evaluations, emphasizing precise measurements. Recommendations include using the Central Limit Theorem, clustering errors, reducing variance, paired-difference analysis, and power analysis for more reliable comparisons and insights.

• Stanford researchers used GPT-4o to simulate over 1,000 individuals' behaviors and personalities with high accuracy, offering alternatives to focus groups but raising ethical concerns about mimicking real people.