The Daily Newsletter for Intellectually Curious Readers

• We scour 100+ sources daily

• Read by CEOs, scientists, business owners and more

• 3.5 million subscribers

Sign up today!

Today:

• Ex-OpenAI Employee Warns AGI

• Lionsgate Partners with Runway AI

• OpenAI o1 AI models for Enterprises Released

• YouTube Integrates Veo into Shorts

• Microsoft, BlackRock Form GAIIP Partnership

• Amazon Launches Project Amelia AI

Ex-OpenAI Employee LEAKED DOC TO CONGRESS!

Former OpenAI employee William Saunders testified before the Senate, warning that the company is closer to creating Artificial General Intelligence (AGI) than most people realize—possibly within three years. AGI refers to highly autonomous systems that can perform most jobs better than humans, potentially disrupting economies and employment worldwide.

Saunders urged for stronger whistleblower protections, independent testing, and government oversight to ensure AI developments are safe and controlled.

WATCH THE VIDEO ON YOUTUBE

Lionsgate Inks Deal With AI Firm to Mine Its Massive Film and TV Library

Lionsgate has teamed up with AI company Runway to develop an exclusive artificial intelligence model trained on Lionsgate's movies and TV shows. This model aims to help filmmakers and creators enhance their projects, making production more efficient and cost-effective. 

While the use of AI offers exciting possibilities for storytelling, it also raises concerns in Hollywood about job losses and the misuse of actors' images. Despite these issues, Lionsgate believes AI can be a valuable tool in content creation. Runway's CEO emphasized their commitment to providing artists with powerful resources to bring their stories to life.

THE HOLLYWOOD REPORTER

OpenAI expands o1 AI models to enterprise and education, competing directly with Anthropic

OpenAI has introduced two new AI models, o1-preview and o1-mini, to ChatGPT Enterprise and Education customers. These models are designed to handle complex reasoning tasks, aiding businesses and educational institutions with advanced coding, scientific research, and intricate problem-solving. 

The o1-preview model excels in solving challenging mathematical and coding problems, outperforming previous models by a significant margin. The smaller o1-mini model is cost-effective and tailored specifically for coding tasks. OpenAI emphasizes safety and ethical use, ensuring customer data is secure and that the models follow guidelines.

VENTUREBEAT

YouTube Shorts to integrate Veo, Google’s AI video model 

YouTube is adding Google's advanced AI video model, Veo, to its Shorts feature. This integration lets creators generate high-quality backgrounds and six-second video clips using simple text prompts. Veo can produce full HD (1080p) videos in various cinematic styles and can edit or remix existing footage. This helps creators add scenes for smoother transitions and better storytelling. 

The new capability will become part of YouTube's Dream Screen later this year. Additionally, YouTube is introducing digital items called "Jewels" for livestream interactions and offering AI tools to help creators brainstorm ideas, create thumbnails, and respond to comments.

TECHCRUNCH

Microsoft, BlackRock form group to raise $100 billion to invest in AI data centers and power

Microsoft and BlackRock are teaming up to form the Global Artificial Intelligence Infrastructure Investment Partnership (GAIIP). This group aims to raise up to $100 billion to build data centers and the energy systems needed for artificial intelligence (AI) technologies. Initially, they plan to gather $30 billion. 

Data centers are facilities that house powerful computers essential for running advanced AI programs, which require a lot of computing power and electricity. By investing in these centers, they hope to support the growing demand for AI applications like chatbots and language models.

CNBC

Amazon debuts Project Amelia, an AI assistant for sellers

Amazon has launched Project Amelia, an AI assistant designed to help sellers grow their businesses on its platform. Built on Amazon's own AI technology, the assistant can answer questions about sales figures and customer traffic, providing summaries and comparisons to previous periods. Sellers can ask simple questions like, "How is my business doing?" and receive detailed insights. 

In the future, Project Amelia will also help resolve issues and manage tasks on behalf of sellers. The goal is to reduce the time and effort sellers spend on managing their online stores, allowing them to focus more on creating great products and satisfying customers.

TECHCRUNCH

🧠RESEARCH

Qwen2.5: A Party of Foundation Models!

Qwen2.5 is a new open-source language model (a program that understands and generates text), plus versions specialized for coding and math. It outperforms earlier models in knowledge, coding, and math tasks. Available in various sizes, it supports many languages and long texts. The team thanks collaborators and plans more improvements.

NVLM: Open Frontier-Class Multimodal LLMs

NVLM 1.0 is a new set of advanced language models that understand both text and images. They match the performance of top models in handling tasks that involve vision and language. By using a better design and high-quality training data, they improve on text tasks too. The models will be shared openly with the community.

OmniGen: Unified Image Generation

OmniGen is a new image generation model that unifies various tasks like text-to-image conversion, image editing, and recognizing edges or human poses into one system. It simplifies the process by eliminating extra components, allowing users to perform complex tasks through simple instructions, making image generation more versatile and user-friendly.

Seed-Music: A Unified Framework for High Quality and Controlled Music Generation

Seed-Music is a new music generation system that creates high-quality songs with precise style control. It uses advanced AI methods to generate vocal music based on inputs like style descriptions, audio references, music scores, and voice prompts. Users can also edit lyrics and melodies directly within the generated audio.

InstantDrag: Improving Interactivity in Drag-based Image Editing

InstantDrag is a fast method for editing images by dragging parts of them, requiring only the image and drag instructions. It uses two special networks to create motion and update the image without extra inputs like masks or text prompts, making real-time, photorealistic editing more interactive and efficient.

🛠️TOP TOOLS

AutoGen - An Open-Source Programming Framework for Agentic AI

Supademo - Close deals faster, drive enablement and scale product onboarding with high-converting, AI-powered product demos.

Guidde - AI platform for business that helps your team create video documentation 11x faster

Cracked - Create animations quickly using text descriptions

MyAudioJournal - Record your thoughts on the go, and our AI will analyze them for you

📲SOCIAL MEDIA

🗞️MORE NEWS

SPECTACLES AR glasses powered by Snap OS

Spectacles '24 are Snap Inc.'s new augmented reality (AR) glasses with see-through lenses, allowing you to view digital enhancements over the real world without obstruction. They aim to bring people together by enabling shared experiences in play, learning, and work, all through an intuitive interface that requires no new skills.

Generate an entire app from a prompt using Together AI’s LlamaCoder

Together AI's LlamaCoder lets users build apps from prompts using Llama 3.1. Over 200,000 apps like quizzes and timers have been made. Developers like that it's open-source—they can use and modify it freely, owning their data without being locked into closed systems.

NVIDIA AI Aerial Launches to Optimize Wireless Networks, Deliver New Generative AI Experiences on One Platform

NVIDIA has introduced AI Aerial, a platform that helps telecom companies enhance wireless networks using artificial intelligence (AI). It allows them to design, simulate, and deploy AI-powered radio access networks (RAN), boosting performance and enabling new services like supporting robots, self-driving cars, and advanced 5G and 6G technologies.

LinkedIn is training AI models on your data

LinkedIn is using users' data to train its AI models without explicit permission. Users are automatically included and must opt out twice—once in settings and once via a form—to prevent their data from being used. Opting out only stops future use, not data already collected.

ClickUp takes on Slack and Teams with its own AI-powered ‘everything’ chat

ClickUp introduced ClickUp Chat, an AI-powered team messaging tool built into its project management platform. It aims to replace apps like Slack and Teams by combining chat, tasks, and projects in one place. Features include AI suggestions, summaries, and turning chat messages into tasks to improve productivity and reduce app-switching.

ByteDance Steps Up AI Chip Efforts

ByteDance, the company behind TikTok, is speeding up plans to create its own artificial intelligence (AI) chips. By collaborating with Taiwan Semiconductor Manufacturing Co., they aim to mass-produce two types of chips by 2026. This could lower their dependence on costly Nvidia chips for AI projects.

This $250 Million Startup’s AI Has Vetted 300,000 Job Candidate

Mercor, started by three 21-year-old Thiel Fellows, uses artificial intelligence to interview and assess job candidates. They've raised $32 million, valuing the company at $250 million. Their AI has conducted 100,000 interviews, evaluating 300,000 candidates, aiming to make hiring more efficient and fair.

Moshi: a speech-text foundation model for real-time dialogue

Moshi is a new speech-to-speech model enabling real-time, two-way voice conversations with minimal delay. Unlike current systems that use separate steps and cause delays, Moshi processes speech directly, handling both the user's and its own speech at the same time. This improves conversation flow and preserves important voice details like emotion.

1X World Model

1X Technologies created a "world model," a computer program that predicts how the world changes in response to a robot's actions. By learning from thousands of hours of robot data, it can simulate complex tasks, helping improve robot training without relying on traditional simulations.