Beat Black Friday with BILL

Get the deal of the year for you and your business when you choose the BILL Divvy Card + expense management software, AND an exclusive gift when you take a demo. Move over, Black Friday.

Choose BILL Spend & Expense to help your business:

• Reap rewards with reliable cash back rates

• Create virtual cards that help protect from fraud & overspending

• Control spending with customizable budget controls

Take a demo by the end of the month and take home a Nintendo Switch, Apple AirPods Pro, Samsung 50" TV, or Xbox Series S—your choice¹.

Request a demo

¹Terms and Conditions apply. See offer page for more details.
BILL Divvy Card is issued by Cross River Bank, Member FDIC, and is not a deposit product.

Today:

• Anthropic Unveils MCP for Integration

• Runway Introduces Frames for Creators

• NVIDIA Launches Fugatto for Audio

• Luma Launches AI Dream Machine

• Intel Nears $8 Billion Grant

Anthropic Unveils MCP for Integration

Anthropic has introduced the Model Context Protocol (MCP), an open-source standard for connecting AI chatbots to data systems. MCP enables two-way integration between AI tools and data sources, simplifying the process for developers by replacing fragmented, custom-built integrations. Companies like Block and Apollo have adopted MCP, and Anthropic offers pre-built connectors for platforms like Google Drive and Slack. 

While MCP promises improved efficiency and context-awareness for AI, its success depends on industry adoption, especially with competition from proprietary solutions like OpenAI's Work with Apps. Anthropic invites developers to collaborate on the protocol, aiming for a unified, scalable AI ecosystem.

Runway Introduces Frames for Creators

Runway has unveiled Frames, a groundbreaking image-generation model that enhances stylistic control and visual quality. Frames allows users to create visually cohesive designs, offering precise control over the aesthetic and atmosphere of generated images. Available through the Gen-3 Alpha and Runway API, Frames supports diverse creative explorations, from retro-inspired styles to dynamic landscapes and portraiture. 

Example themes include vintage album art, Japanese zine aesthetics, and experimental collages. Pre-installed safety features ensure responsible use. This release reflects Runway’s commitment to innovation in creative tools, as demonstrated through partnerships with industry leaders like Lionsgate and Tribeca Festival. Frames aims to redefine visual storytelling.

NVIDIA Launches Fugatto for Audio

NVIDIA introduced Fugatto, a powerful AI model for audio generation and transformation, combining music, voices, and sounds based on text or audio prompts. It can prototype songs, alter accents and emotions, or create unique soundscapes like thunderstorms merging into bird songs. 

Fugatto offers fine control over artistic elements, such as blending accents and emotions or evolving sounds over time. With 2.5 billion parameters, it was developed using NVIDIA DGX systems and extensive datasets. This tool aims to revolutionize music, gaming, advertising, and education, enabling creativity with unprecedented flexibility and realism in sound design.

Luma Launches AI Dream Machine

Luma unveiled Dream Machine, an innovative AI-powered platform that generates stunning images and videos effortlessly. With Photon, its advanced image model, users can ideate, create, and refine visuals quickly and intuitively using simple text prompts. The tool supports customizing styles, modifying existing visuals, and generating consistent characters across diverse settings. 

Features like Reference and Remix and Brainstorm allow for unlimited creative exploration, while its efficiency ensures high-resolution outputs with 8x speed compared to similar tools. Dream Machine empowers creators in fields like design, music, and film to visualize and share their ideas seamlessly on web and iOS platforms.

Intel Nears $8 Billion Grant

Intel is nearing a deal with the Commerce Department for an $8 billion CHIPS Act grant to expand chip manufacturing. This funding, expected to be finalized soon, aims to support Intel's struggling business, which has faced significant losses and a potential takeover bid by Qualcomm. 

Despite frustrations over delays in CHIPS Act grants, Intel CEO Pat Gelsinger remains optimistic about the legislation’s job-creation potential. However, a $500 million reduction in the grant reflects concerns about Intel's ability to meet commitments. Alongside this grant, Intel also secured a $3 billion Department of Defense contract to manufacture chips, offering some financial relief.

🧠RESEARCH

TÜLU 3: Pushing Frontiers in Open Language Model Post-Training

The TÜLU 3 project presents open-source advanced post-trained language models, outperforming top models like Llama 3.1 and GPT-4o-mini. It provides detailed training recipes, datasets, and evaluation tools for reproducibility. Key techniques include supervised finetuning, Direct Preference Optimization, and a novel Reinforcement Learning method, offering transparency and accessibility in refining AI language skills.

BALROG: Benchmarking Agentic LLM and VLM Reasoning On Games

The BALROG benchmark evaluates Large Language and Vision Language Models (LLMs and VLMs) on reasoning and planning through challenging games. It highlights strengths in simple tasks but reveals major gaps in complex, vision-based decision-making. BALROG provides diverse environments, fine-tuned metrics, and open resources to advance research in dynamic, agent-driven AI capabilities.

VideoEspresso: A Large-Scale Chain-of-Thought Dataset for Fine-Grained Video Reasoning via Core Frame Selection

The VideoEspresso dataset advances video reasoning by providing high-quality VideoQA pairs with detailed spatial and temporal coherence. Using semantic-aware methods and GPT-4o-generated annotations, it reduces redundancy and enriches reasoning through Chain-of-Thought (CoT) processes. A Hybrid LVLMs framework further enhances performance, outperforming baselines in video reasoning tasks.

VideoRepair: Improving Text-to-Video Generation via Misalignment Evaluation and Localized Refinement

VideoRepair enhances text-to-video generation by addressing misalignments between text prompts and video outputs, particularly for complex scenes. This training-free framework detects inaccuracies, plans refinements, and adjusts specific regions while preserving correct areas. It significantly improves alignment metrics on benchmarks, offering a practical solution for more accurate and coherent video generation.

MyTimeMachine: Personalized Facial Age Transformation

MyTimeMachine creates personalized facial age transformations by combining global aging patterns with individual photo collections. Using as few as 50 images, it employs an Adapter Network and specialized loss functions to generate realistic, identity-preserving aging effects. This method extends to videos, achieving superior temporal consistency and accurate age predictions compared to existing techniques.

🛠️TOP TOOLS

Cypherscribe - Connect your data, customize, and get an interactive full stack web-app in 18 seconds without a single line of code

Demodesk - AI that summarizes your calls, automates CRM management, and delivers personalized, actionable sales coaching.Maxim AI - End-to-end AI evaluation and observability platform, empowering modern AI teams to ship products with quality, reliability, and speed

Moonbeam - Your long-form writing AI assistant

Boltic - Enterprise-Grade Platform for AI-Powered Workflow & Automation

MyAudioJournal - Talk, train your mind, and get actionable insights to achieve your long-term goals—all in complete privacy.

📲SOCIAL MEDIA

🗞️MORE NEWS

• Quartr partners with Perplexity to integrate live earnings call transcripts and AI-powered summaries, making advanced financial insights accessible to all users, including retail investors, through Perplexity’s innovative search engine and new finance widget.

• Amazon is developing AI chips to reduce reliance on Nvidia, targeting affordability and performance. Their new Trainium2 chip aims to power AI workloads, but success hinges on simplifying software and meeting customer needs in a competitive market.

• Thomson Reuters integrates OpenAI's o1-mini model into its CoCounsel legal AI assistant, enhancing precision in legal document analysis and task management. This specialized approach redefines enterprise AI with tailored, high-performance models for legal workflows.

• PlayAI, formerly PlayHT, offers advanced voice cloning and text-to-speech solutions, including tools for creating audio content and automating tasks like customer calls. Despite ethical concerns and competition, it raised $20M to enhance its AI voice models and agent platform.