The fastest way to build AI apps

• Writer Framework: build Python apps with drag-and-drop UI

• API and SDKs to integrate into your codebase

• Intuitive no-code tools for business users

Start building with AI Studio

Today:

• ChatGPT App for Windows

• ChatGPT Targets Businesses with Bain

• Google Launches NotebookLM for Businesses

• Blumhouse Partners with Meta Movie Gen

• Nvidia's AI Crushes GPT-4 Benchmarks

ChatGPT App for Windows

OpenAI is testing a new ChatGPT app for Windows, but it’s only available to paid users. The app, which can be downloaded from the Microsoft Store, lets users ask the AI questions in a dedicated window, and access it quickly using a keyboard shortcut. It also allows uploading files and photos, and offers a preview of OpenAI's new reasoning model. 

However, some features, like advanced voice mode, are not yet available. OpenAI plans to release the app to all users later this year. The company has addressed previous security issues by encrypting stored data.

ChatGPT Targets Businesses with Bain

OpenAI and Bain & Co. are expanding their partnership to bring AI tools, including ChatGPT, to businesses. Bain’s 13,000 consultants now use ChatGPT Enterprise, targeting industries like retail and life sciences with customized AI solutions. This collaboration aims to provide industry-specific tools for clients, such as AI-based planning tools for retailers and document automation for life sciences. 

OpenAI currently has over one million business customers, and its usage of APIs has doubled since July 2024. The partnership helps OpenAI tap into larger business budgets as the company scales its operations and expands its reach across sectors.

Google Launches NotebookLM for Businesses

Google is launching a paid version of its AI tool, NotebookLM, aimed at businesses. Called NotebookLM Business, it offers features designed for companies, universities, and organizations, with a focus on boosting productivity and collaboration. The tool, currently in a pilot program, allows users to upload materials into "notebooks" and ask the AI questions, while new features include Audio Overview, which generates narrated research summaries. 

Google plans to announce general availability and pricing later this year. Enhanced data privacy and security are included, along with options to customize and share notebooks with team members.

Blumhouse Partners with Meta Movie Gen

Meta has launched "Meta Movie Gen," a suite of AI models that allows users to create custom videos and audio using simple text inputs. This tool, which generates high-quality 1080p HD videos with sound, is currently in a pilot program with feedback from filmmakers like Aneesh Chaganty and Casey Affleck. 

In partnership with Blumhouse, Meta is refining the technology to support the creative community. The AI models offer new possibilities for visual and audio creation, aiming to inspire and assist filmmakers. Meta plans to expand the program and release the tool to the public in 2025.

Nvidia's AI Crushes GPT-4 Benchmarks

Nvidia has quietly launched a new AI model, Llama-3.1-Nemotron-70B-Instruct, which surpasses top models like OpenAI's GPT-4 in performance benchmarks. The model was introduced on Hugging Face and is notable for its ability to handle complex queries with accuracy and without extra prompts. 

This move highlights Nvidia's shift from its traditional GPU dominance to high-performance AI software, positioning it as a major competitor in the AI space. Offering free access through an API, Nvidia aims to make the model accessible to businesses, but it cautions that the model isn’t yet optimized for specialized fields like math or legal reasoning.

🧠RESEARCH

VidEgoThink: Assessing Egocentric Video Understanding Capabilities for Embodied AI

VidEgoThink, a benchmark for evaluating how well AI models understand egocentric (first-person) video. It focuses on four tasks: video question-answering, planning, visual grounding, and reward modeling. Despite advancements, current models, including GPT-4o, struggle with these tasks, highlighting the need for further improvement in AI’s real-world, first-person applications.

HumanEval-V: Evaluating Visual Understanding and Reasoning Abilities of Large Multimodal Models Through Coding Tasks

HumanEval-V, a benchmark to evaluate how well multimodal AI models (LMMs) understand visual information and solve coding tasks. It includes 108 Python tasks with visual contexts. Results show that even advanced models like GPT-4o struggle, highlighting the need for improvement in LMMs' visual reasoning and coding abilities.

The Curse of Multi-Modalities: Evaluating Hallucinations of Large Multimodal Models across Language, Visual, and Audio

The paper examines hallucinations in large multimodal models (LMMs), where outputs do not match multimodal inputs (language, visual, audio). It identifies two causes: overreliance on one modality and incorrect correlations between modalities. The proposed CMM benchmark evaluates these hallucinations, revealing integration imbalances and biases, suggesting improvements for more accurate and reliable LMMs.

Large Language Model Evaluation via Matrix Nuclear-Norm

Matrix Nuclear-Norm as a faster and efficient method to evaluate large language models (LLMs), focusing on data compression and reducing redundancy. It reduces computational complexity compared to traditional methods like Matrix Entropy. Tests show that it significantly speeds up performance, especially for larger models, while maintaining accuracy.

DocLayout-YOLO: Enhancing Document Layout Analysis through Diverse Synthetic Data and Global-to-Local Adaptive Perception

DocLayout-YOLO, a method for improving document layout analysis by balancing speed and accuracy. It uses the DocSynth-300K dataset for pre-training, enhancing document recognition. A new Global-to-Local Controllable Receptive Module allows better handling of varying document elements. DocLayout-YOLO outperforms existing methods in both speed and accuracy, validated through experiments.

🛠️TOP TOOLS

Replit - AI-powered software development & deployment platform for building, sharing, and shipping software fast.

Concourse - First AI Analyst for corporate finance teams

Langflow - The easiest way to create and share AI-driven apps

Suno Scenes - Create unique songs from your favorite photos and videos - all from your phone.

Hailuo AI - AI video model that generates high-quality 6-second videos from text prompts

📲SOCIAL MEDIA

🗞️MORE NEWS

• ChatGPT's web traffic surged 112% year-over-year, reaching 3.1 billion visits in September 2024. This growth marks a significant recovery from 2023's summer slump. The app's user base has expanded rapidly, especially in the US and UK.

• Perplexity's new Internal Knowledge Search lets enterprises search both internal files and the web in one platform. Users can upload key documents, enhancing productivity by combining internal research with external insights.

• Modders are using AI to enhance games like Stardew Valley and Skyrim, enabling more dynamic conversations with NPCs. These AI-powered mods use OpenAI’s API, allowing players to engage in seemingly endless dialogues, but come with limitations such as repetitive personalities and cost barriers for players. 

• Pika Labs’ AI-powered video platform has launched new special effects like crumble, dissolve, and deflate, enhancing creative video manipulation. The platform offers various subscription plans, allowing users to apply effects easily and explore innovative content solutions.

• Google Cloud's Vertex AI platform, now available for healthcare, enhances data query capabilities with generative AI tools like MedLM. It aims to reduce healthcare workers' administrative burden, helping them focus more on patient care.

• Toyota and Hyundai's Boston Dynamics have partnered to accelerate the development of AI-powered humanoid robots. Combining Toyota's machine learning expertise and Boston Dynamics' Atlas robot, they aim to enhance human-robot interactions and explore new AI-driven use cases.

• Meta is facing criticism from the Open Source Initiative (OSI) for misusing the term "open-source" to describe its Llama AI models, which restrict full transparency and usage. The OSI argues that Meta's actions confuse users and harm open-source innovation by limiting experimentation and control over the models. 

• Salesforce CEO Marc Benioff criticized Microsoft's AI assistant, Copilot, calling it "Clippy 2.0" and saying it lacks accuracy. While the critique targets a rival, Benioff also expressed skepticism about AI's broader potential, contrasting with his excitement about Salesforce’s AI tools like Agentforce.