Receive Honest News Today

Join over 4 million Americans who start their day with 1440 – your daily digest for unbiased, fact-centric news. From politics to sports, we cover it all by analyzing over 100 sources. Our concise, 5-minute read lands in your inbox each morning at no cost. Experience news without the noise; let 1440 help you make up your own mind. Sign up now and invite your friends and family to be part of the informed.

Sign up today!

Today:

• Clickative AI Powers Video Upscaling

• OpenAI Joins Forces with Anduril

• Colossus Supercomputer to Expand Tenfold

• DeepMind AI Beats Weather Forecasting System

• Meta Invests $10 Billion in Louisiana

AI Video Just Went TOO FAR... AI Video Upscaler

AI video generation is advancing rapidly, with rumors of OpenAI releasing a new AI video model soon. Tencent, a Chinese company, is also preparing to release an open-source AI video generation tool. 

Meanwhile, AI-powered video upscaling is emerging, following the success of Magnific, an AI image upscaler. Sergio Suave’s Clickative AI, powered by Magnific’s technology, aims to upscale videos, improving details and textures. Although challenges like character consistency remain, the technology is progressing. Clickative AI is still in beta, but a sign-up sheet for early testers is available, likely requiring Magnific credits to use.

WATCH THE VIDEO ON YOUTUBE

OpenAI Joins Forces with Anduril

OpenAI and Anduril Industries, a defense technology company, announced a partnership to develop AI solutions for national security. The collaboration focuses on improving the U.S. military's ability to detect and counter aerial threats, such as drones, in real time. By training AI models using Anduril's data on counter-unmanned aircraft systems (CUAS), the partnership aims to enhance defense capabilities. 

OpenAI's CEO, Sam Altman, emphasized that the technology will protect U.S. military personnel. The move comes amid global efforts to develop AI-powered weapons and autonomous military systems.

Colossus Supercomputer to Expand Tenfold

Elon Musk’s xAI plans to expand its Colossus supercomputer by ten times, adding over 1 million GPUs. The facility, already the world’s largest, will grow in Memphis, Tennessee. Colossus powers Musk's chatbot, Grok, which aims to compete with ChatGPT. The expansion, requiring billions in investment, involves building vast infrastructure and acquiring expensive GPUs. 

Musk is pursuing this project independently, unlike rivals like OpenAI, which rely on partnerships for computing power. Despite limited commercial offerings, Colossus' rapid development in just three months has earned praise, making it one of the most powerful supercomputers globally.

DeepMind AI Beats Weather Forecasting System

Google DeepMind's new AI weather model, GenCast, outperforms the best current forecasting system, predicting weather up to 15 days ahead in just minutes. Unlike traditional models, which rely on complex physics, GenCast uses historical weather data to identify patterns, making it faster and more accurate, especially for extreme weather events like hurricanes. It produces probabilistic forecasts, offering not only predictions but also the likelihood of accuracy. 

Trained on global data from 1979-2018, it is more reliable than existing models and can generate forecasts much faster. The code is publicly available, encouraging further advancements in weather prediction technology.

Meta Invests $10 Billion in Louisiana

Governor Jeff Landry announced that Meta will invest $10 billion in a new artificial intelligence (AI) data center in northeast Louisiana. The project, expected to create over 500 direct jobs and 1,000 indirect jobs, will occupy 4 million square feet on the 2,250-acre Franklin Farm site. 

Meta plans to match its energy use with 100% renewable power and invest over $200 million in local infrastructure. The development will boost the state’s tech sector, create new workforce opportunities, and promote local education in digital skills. The center's construction is set to continue until 2030, marking a significant economic milestone for Louisiana.

🧠RESEARCH

VideoGen-of-Thought: A Collaborative Framework for Multi-Shot Video Generation

VideoGen-of-Thought (VGoT) is a new system designed to generate multi-shot, movie-like videos. It divides the process into four parts: script generation, keyframe creation, shot-level video generation, and smoothing for consistency. VGoT ensures logical storytelling and visual coherence across shots, outperforming existing video generation models.

MALT: Improving Reasoning with Multi-Agent LLM Training

MALT introduces a method for training multiple large language models (LLMs) to collaborate on reasoning tasks. It uses a sequential setup with specialized roles (generator, verifier, and refiner) and a synthetic data generation process. This approach improves problem-solving performance, showing notable gains in mathematical and common sense reasoning tasks.

Motion Prompting: Controlling Video Generation with Motion Trajectories

Motion Prompting introduces a new method for controlling video generation using motion trajectories, offering more precise control over dynamic actions and temporal elements than text prompts. This approach allows for flexible, user-defined motion, including camera and object movements, with applications in image interaction and editing, showcasing realistic physics and strong performance.

o1-Coder: an o1 Replication for Coding

O1-CODER replicates OpenAI's o1 model for coding tasks, integrating reinforcement learning (RL) and Monte Carlo Tree Search (MCTS) to improve reasoning. It trains a Test Case Generator for standardized code testing, generates pseudocode, and refines the model to produce full code. The report explores challenges in applying o1-like models to real-world tasks.

VideoLights: Feature Refinement and Cross-Task Alignment Transformer for Joint Video Highlight Detection and Moment Retrieval

VideoLights introduces a novel framework for video highlight detection and moment retrieval, improving video-text alignment and task integration. It uses feature refinement, bi-directional cross-modal fusion, and task feedback mechanisms to strengthen video-text representation. The approach, enhanced by large vision-language models, achieves state-of-the-art results on several benchmarks.

🛠️TOP TOOLS

GemPages - Build Sales Funnels with Post-Purchase Upsell Offers

ElevenLabs GenFM - Tune in as AI co-hosts generate smart podcasts from any of your PDFs, articles, ebooks and more.

Ubique - Record a single video massage and generate 1000+ personalized videos.

Dripify - The multifunctional LinkedIn automation tool designed to help your sales team improve LinkedIn prospecting and close more deals — all on complete autopilot

📲SOCIAL MEDIA

🗞️MORE NEWS

• OpenAI CEO Sam Altman suggests artificial general intelligence might arrive sooner than expected but won't drastically change the world. He downplays previous claims, predicting gradual economic impacts and a long path to superintelligence.

• DeepMind's Genie 2 creates interactive 3D worlds from text and images, simulating actions and environments like video games. While aimed at research and prototyping, it can generate diverse worlds with realistic physics, lighting, and animations.

• Three leaders from Google’s viral NotebookLM project, including the team lead, are leaving to start their own stealth startup. The new company plans to create user-friendly AI products but hasn’t revealed its focus or name yet.

• Amazon unveiled new AI models at its AWS conference, competing with Adobe and Meta. The models include tools for generating text, images, and videos, aiming to offer faster, cheaper, and more customizable AI solutions for businesses.

• Spotify's latest feature for 2024 Wrapped has added a personalized, AI-generated podcast that analyzes your listening habits. Powered by Google’s NotebookLM, this feature includes two synthetic hosts discussing your top tracks, artists, and habits in a conversational style.