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:

• Palantir, Anthropic Partner with AWS

• Jarvis AI Accidentally Goes Public

• Mistral AI Challenges OpenAI's Lead

• ByteDance Launches Expressive Portrait Animation

• Gemini Models Boost AI Coding Assistants

Palantir, Anthropic Partner with AWS

Palantir and Anthropic have partnered with Amazon Web Services (AWS) to deliver Claude AI models to U.S. defense and intelligence agencies. This collaboration allows secure, rapid data analysis within Palantir’s AI Platform (AIP) on AWS, facilitating real-time decision-making and operational efficiency for the government. By leveraging Claude's AI capabilities, these agencies can handle complex data faster, identify trends, and streamline tasks like document review. 

Palantir's setup, certified for strict security standards, uses AWS technology, including Amazon SageMaker, for advanced, responsible AI deployment in critical missions. This move extends Palantir's commercial success with Claude to secure government environments.

Jarvis AI Accidentally Goes Public

Google’s new AI assistant, Jarvis, briefly appeared on the Chrome Web Store, giving users a sneak peek of its capabilities before its planned December launch. Designed to handle tasks like web surfing, shopping, and booking, Jarvis can operate a computer without direct human input. 

However, due to restricted permissions, users couldn’t fully test its functions before Google removed the prototype. Jarvis will enter a competitive field, facing Anthropic’s Claude AI, which can also control computers by typing, activating buttons, and moving the cursor.

Mistral AI Challenges OpenAI's Lead

French startup Mistral AI launched a new content moderation API, aiming to compete with OpenAI by tackling harmful content in 11 languages. Using its Ministral 8B model, the API identifies harmful content across nine categories, including hate speech, violence, and personal information. This tool supports Mistral’s own Le Chat platform and offers multilingual capabilities, giving it an edge over competitors focused primarily on English. 

The API, already adopted by major partners like SAP, emphasizes safety, privacy, and compliance with European regulations. Mistral’s dual approach to on-device AI and content moderation makes it a promising player in enterprise AI.

ByteDance Launches Expressive Portrait Animation

X-Portrait 2 is an advanced technology developed by ByteDance and Tsinghua University that creates realistic and expressive animated videos from a single photo. Users provide a still portrait and a video showing facial expressions. The system then mimics the expressions from the video onto the photo, producing dynamic animations without the need for complex equipment. 

X-Portrait 2 uses a highly detailed expression model and powerful image-generation techniques to capture subtle facial movements like pouting or frowning. It separates the person's appearance from their movements, allowing the technology to work with both real and cartoon images. This makes it useful for storytelling, character animation, virtual assistants, and special effects in movies.

Gemini Models Boost AI Coding Assistants

Google has introduced a collaboration with Sourcegraph to enhance AI coding assistants using its Gemini models with long-context capabilities. These models, tested with Sourcegraph’s Cody assistant, improve code understanding in large, complex codebases by expanding context windows, allowing AI to generate more accurate and concise responses. 

In testing, Cody’s performance improved in key benchmarks—Essential Recall, Concision, and Helpfulness—while reducing AI "hallucinations" (false information) significantly. Despite increased response times with longer contexts, Sourcegraph optimized this with caching, reducing wait times to around five seconds. 

🧠RESEARCH

Both Text and Images Leaked! A Systematic Analysis of Multimodal LLM Data Contamination

This study examines data contamination in multimodal large language models (MLLMs), which mix text and images. Traditional detection methods struggle with MLLMs’ complex training. The researchers developed MM-Detect, a tool to spot contamination in MLLMs, revealing performance boosts from leaked data. They highlight risks across pre-training and fine-tuning phases.

A Comprehensive Survey of Small Language Models in the Era of Large Language Models: Techniques, Enhancements, Applications, Collaboration with LLMs, and Trustworthiness

This survey highlights the rise of Small Language Models (SLMs) as efficient alternatives to Large Language Models (LLMs), which have high computational costs and limited domain-specific expertise. SLMs offer faster, cost-effective solutions tailored for specific fields, like healthcare, and resource-limited settings. The study organizes existing SLM techniques, applications, and improvements, providing a standardized framework and GitHub resources.

Large Language Models Orchestrating Structured Reasoning Achieve Kaggle Grandmaster Level

Agent K v1.0 is an autonomous data science agent designed to fully automate and optimize complex tasks across various data types. Using structured reasoning and experience-based learning, it completes Kaggle competitions independently, achieving results comparable to human Grandmasters. With a 92.5% task success rate, Agent K demonstrates high proficiency, earning multiple Kaggle medals.

How Transformers Solve Propositional Logic Problems: A Mechanistic Analysis

This study explores how transformers solve complex propositional logic problems, focusing on their internal reasoning mechanisms. Researchers trained a small, three-layer transformer to tackle a challenging logic task, uncovering specific "planning" and "reasoning" circuits within its attention blocks. Expanding to a larger model, Mistral 7B, they identified key components enabling logical reasoning, offering insights into transformers' planning and reasoning abilities.

Polynomial Composition Activations: Unleashing the Dynamics of Large Language Models

"PolyCom," a new type of activation function for transformers that enhances their ability to handle complex data patterns. PolyCom boosts transformer dynamics by capturing higher-order interactions, leading to better accuracy and faster convergence. Tested on large language models, PolyCom demonstrated significant improvements over traditional activation functions.

🛠️TOP TOOLS

Genmo Playground - An open-source video generation model 

Poppop AI - Remove vocals or instrumantals from any song

Sona - Captures your conversations and provides insights that matter most to you.

Video Ocean - Create videos from text and images

Hume - Combine voices and personalities generated by our speech-language model, EVI 2, with supplemental LLMs and tools like the new Claude 3.5 Haiku

📲SOCIAL MEDIA

🗞️MORE NEWS

• Microsoft is bundling its AI-powered Copilot features into Microsoft 365 Personal and Family subscriptions, initially in select Asia-Pacific markets. While prices are increasing, the change removes the previous $20 Copilot Pro fee, expanding AI access and testing future pricing adjustments globally.

• Apple’s iOS 18.2 beta includes a ChatGPT integration within Siri, setting daily usage limits for advanced AI capabilities. Users can bypass these limits by subscribing to ChatGPT Plus, providing expanded features for enhanced Siri responses.

• Nous Research has launched Nous Chat, its first public chatbot, powered by its Hermes 3-70B model. Designed with customizable prompt suggestions and guardrails, Nous Chat offers a user-friendly experience, though it lacks some advanced features seen in other chatbots.

• Donald Trump plans to repeal President Biden’s 2023 AI safeguards, including dismantling the US AI Safety Institute and implementing steep tariffs on GPU imports. This deregulation could shift AI oversight to state governments, with potential impacts on tech innovation and trade.