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:

• Sam Altman Reveals AI Breakthroughs

• Sam Altman Joins Lurie’s Team

• Mistral AI Enhances le Chat

• Qwen2.5 Redefines Long-Context AI

• ElevenLabs Introduces Conversational AI Platform

• Ray-Ban Glasses Now in Europe

Sam Altman "Level 4 Innovator is MUCH closer" | PLUS "Unsupervised Sentiment Neuron" breakthrough

Sam Altman, OpenAI’s CEO, spoke with Y Combinator's Gary Tan, highlighting key milestones and challenges in AI development. OpenAI’s journey began with skepticism, focusing on AGI despite critics. Altman emphasized unwavering conviction, computing power, and risks in scaling AI. 

A pivotal moment was discovering the "unsupervised sentiment neuron," sparking the GPT series, transforming AI into tools with real-world impact. The discussion also explored AI's leap from reasoning (Level 2) to innovation and autonomy (Levels 3-5), showcasing breakthroughs like wildfire prediction and encryption algorithms. Altman warned about safety risks, urging proactive measures as AI accelerates rapidly.

WATCH THE VIDEO ON YOUTUBE

Sam Altman Joins Lurie’s Team

San Francisco mayor-elect Daniel Lurie has named OpenAI CEO Sam Altman as a co-chair for his transition team, signaling a collaboration with tech leaders to tackle city issues. Lurie, a political newcomer, invested $9 million in his campaign, defeating incumbent London Breed. 

Altman will guide innovation strategies, rebuild relationships with tech figures, and help address public safety concerns that have driven many professionals away. Other team members include ex-Twitter CFO Ned Segal and community leaders. Lurie aims to attract young talent, boost startups, and restore confidence in city leadership, with Altman positioned to influence both San Francisco and tech politics.

Mistral AI Enhances le Chat

Mistral AI has introduced updates to its AI assistant, le Chat, enhancing productivity and creativity for users. Key features include web search with citations, a collaborative Canvas interface for ideation, advanced document/image understanding powered by the multimodal Pixtral Large model, and seamless image generation through Black Forest Labs. Users can automate repetitive workflows with task agents. 

Le Chat's free beta offers unmatched versatility, from research to creative tasks, with tools like inline editing and versioning. Additionally, Pixtral Large excels in document analysis, multilingual OCR, and reasoning, outperforming competitors. Mistral AI prioritizes accessible, cutting-edge AI tools for personal and professional growth.

Qwen2.5 Redefines Long-Context AI

The Qwen team has unveiled Qwen2.5-Turbo, extending context length to 1 million tokens for enhanced processing of long texts, like novels, transcripts, or large codebases. Achieving 93.1 on the RULER benchmark, it outperforms GPT-4 in long-context tasks. Despite this expansion, short-text capabilities remain uncompromised. Using sparse attention, inference speeds improved by up to 4.3x, reducing processing time for 1M tokens to 68 seconds. Cost efficiency is maintained at ¥0.3/1M tokens. 

Available via API, Qwen2.5-Turbo excels in tasks like summarizing novels or analyzing repositories. Future updates aim to refine long-sequence performance and enhance inference efficiency for real-world applications.

ElevenLabs Introduces Conversational AI Platform

ElevenLabs has launched a platform for building conversational AI agents, offering customizable features like tone, response length, and integration of knowledge bases. Users can personalize agents with system prompts, voice settings, and large language models like Gemini or GPT. The platform supports Python, JavaScript, and more, allowing developers to create advanced bots tailored to specific needs. Key features include data collection criteria, speech-to-text integration, and seamless customization via APIs. 

Competing with OpenAI’s real-time conversational API, ElevenLabs emphasizes flexibility and model-switching capabilities. Aiming for a $3 billion valuation, the company positions itself as a leader in conversational AI innovation.

Ray-Ban Glasses Now in Europe

Meta is expanding the availability of Meta AI on Ray-Ban Meta glasses to France, Italy, Ireland, and Spain. Users can now interact hands-free with Meta AI in French, Italian, and Spanish to get answers, recommendations, or creative ideas while on the go. Features allowing AI to provide information about visible objects remain limited to the US, Canada, and Australia. 

Since its 2023 launch, Meta has ensured compliance with European regulations and plans to roll out additional features and expand to more countries. This marks a significant step in enhancing smart glasses' utility across Europe.

🧠RESEARCH

LLaVA-o1: Let Vision Language Models Reason Step-by-Step

LLaVA-o1 is a new vision-language model designed for better reasoning in visual question-answering tasks. Using structured, step-by-step methods and a small dataset, it outperforms larger models like GPT-4o-mini, achieving improved precision and efficiency.

Region-Aware Text-to-Image Generation via Hard Binding and Soft Refinement

RAG introduces a novel method for region-aware text-to-image generation, combining precise region binding with smooth refinement for detailed layouts. It enables editing specific areas without inpainting models, improving control, adaptability, and prompt accuracy over prior methods.

GaussianAnything: Interactive Point Cloud Latent Diffusion for 3D Generation

GaussianAnything is a 3D generation framework using a point cloud latent space for scalable, high-quality outputs. It supports diverse inputs like text, images, and captions while enabling 3D editing and outperforming existing methods in detail and flexibility.

Xmodel-1.5: An 1B-scale Multilingual LLM

Xmodel-1.5 is a 1-billion-parameter multilingual AI model trained on 2 trillion tokens, excelling in Thai, Arabic, French, Chinese, and English. It advances multilingual AI research with open-source models, code, and a Thai evaluation dataset.

The Dawn of GUI Agent: A Preliminary Case Study with Claude 3.5 Computer Use

This study examines Claude 3.5 Computer Use, an AI model for GUI-based tasks, showcasing its ability to handle language-to-desktop actions. It highlights strengths, limitations, and offers tools for GUI automation, inspiring future research.

🛠️TOP TOOLS

AI Game Master - Pick your plot, battle it out with cool text moves, and steer your story wherever you want. 

Tavrn - AI-powered medical chronologies for attorneys.

Sharbo AI - Analyze, compare, and track competitor features relative to your product.

Integry - App Functions for AI

Recall.ai Output Media API - Generate and stream low-latency audio and video directly into a video conference

📲SOCIAL MEDIA

🗞️MORE NEWS

• Perplexity’s AI search engine now lets Pro subscribers buy products directly with a “Buy with Pro” button, offering free shipping and streamlined purchasing. New tools include image-based shopping and expanded features for merchants.

• Former Google employees launched TwinMind, an AI app that remembers users’ activities, offering personalized assistance. Valued at $30M, it transcribes audio, integrates with calendars, and drafts emails, prioritizing privacy and efficiency.

• Google.org has launched a $20 million fund to support AI-driven scientific breakthroughs. The funding will aid global nonprofits and academics tackling challenges in biology, sustainability, and diseases, while providing cloud credits and expert assistance.

• ESPN is testing "FACTS," an AI avatar for SEC football broadcasts, to present complex stats alongside analysts. It aims to enhance fan engagement without replacing journalists, emphasizing innovation in sports coverage.

• Researchers at Washington State University developed an AI model that analyzes tissue images faster and more accurately than humans, revolutionizing disease diagnostics by identifying pathologies missed by experts in weeks instead of years.

• Nvidia is collaborating with Google Quantum AI to simulate quantum processor designs using its Eos supercomputer and CUDA-Q platform. This partnership enables faster, cost-effective simulations to address quantum noise, advancing scalable, practical quantum computing solutions.

• Scientists developed Evo, an AI model trained on microbial genomes, capable of predicting genetic mutation effects and generating new DNA sequences. It advances genome-scale research but raises ethical and safety considerations for future applications.