Welcome to The AI Entrepreneurs, where we’re diving into the most transformative AI advancements of the year! From healthcare breakthroughs with NVIDIA and Google Research to the debut of OpenAI’s lightning-fast image generation model, this issue covers the powerful ways AI is reshaping our world. Discover Meta's move to bring real-time Reuters news to its chatbot, Cohere's new multimodal search capabilities, and more. Dive in for the latest insights driving the AI frontier forward!

AI in Healthcare 🏥

AI is transforming healthcare in incredible ways!

Dive into the latest issue of AIHealthTech Insider for all the details! 🔽

In this issue, we explore new AI advancements in healthcare. NVIDIA and Deloitte have developed AI virtual assistants to help calm patients before surgery, and Google Research has enhanced 3D medical imaging with their CT Foundation. These technologies are improving patient care. We also discuss AI innovations presented at HLTH 2024 and UCLA's SLIViT model, which uses NVIDIA technology to make 3D medical image analysis and disease detection faster.

Rumor Control 🚫

Altman Shuts Down Rumors of New AI Model "Orion"

OpenAI CEO Sam Altman has firmly dismissed recent rumors about a new AI model called "Orion" supposedly launching in December. The report from The Verge, citing anonymous sources, claimed that OpenAI was preparing to release a model up to 100 times more powerful than GPT-4, with Microsoft’s engineers allegedly integrating it by November. Altman took to X (formerly Twitter) to refute the story, calling it “pure fantasy” and criticizing the media's willingness to “print random fantasy.”

Altman's post dismissed rumors about OpenAI's next AI model, Orion, suggesting media reports are inaccurate. He hinted at future advancements, saying, “there’s plenty of great stuff coming your way,” but Orion may not be one of them. This highlights tensions between tech companies and media, leaving the industry guessing about OpenAI's actual plans, which may include significant developments not under the "Orion" name.

AI in Image Generation 🎉

OpenAI’s New sCM Model Generates Images 50x Faster!

OpenAI has unveiled a breakthrough method called simplified, stabilized, and scaled Consistency Models (sCM), aimed at streamlining and accelerating the training of AI image generation models. Building on previous research with Consistency Models (CMs), this new technique optimizes fast image sampling, allowing for high-quality image generation in only two computational steps—a dramatic improvement over traditional diffusion-based models that require many more steps.

The sCM method enhances model training by improving stability and scalability. OpenAI's largest sCM model, with 1.5 billion parameters, generates images in 0.11 seconds on an A100 GPU, 50 times faster than previous models. It achieved FID scores of 2.06 on CIFAR-10 and 1.88 on ImageNet, indicating high image quality with minimal computational cost. The sCM approach scales effectively, with image quality improving as model size increases, suggesting potential applications in video, audio, and 3D model generation.

Quantized Llama Models 📱

Meta Brings Faster, Lighter AI Models to Mobile Devices

Meta has introduced its first quantized Llama models, designed to run efficiently on mobile devices while offering faster performance and reduced memory usage. The 1B and 3B Llama models, built using Quantization-Aware Training (QAT) and SpinQuant, deliver a 2-4x speedup and a 41% reduction in memory usage compared to the original models. These innovations allow developers to deploy on-device AI without needing significant compute resources.

In collaboration with Qualcomm and MediaTek, Meta optimized these models for ARM CPUs with plans to further enhance performance using NPUs. This development makes the models not only faster but also more accessible for real-time AI tasks on mobile platforms, catering to developers looking for lightweight, powerful solutions.

By enabling on-device AI with reduced footprint and improved performance, Meta is paving the way for more privacy-conscious AI applications and enhancing mobile AI experiences across industries like healthcare, gaming, and real-time communication.

AI and News 🗞️

Meta Partners with Reuters for Real-Time News in Chatbot

Meta has secured a multi-year agreement with Reuters to integrate real-time news content into its Meta AI chatbot, marking the tech giant's first formal news partnership in the AI realm, according to sources who spoke with Axios. This collaboration enables users in the U.S. to access Reuters’ news updates through Meta's chatbot on platforms like Facebook, Instagram, WhatsApp, and Messenger, where responses will feature summaries and direct links to Reuters articles.

Starting Friday, users can access up-to-date news by asking questions on current events, with Reuters compensated for content use. While Meta's AI has focused on creative and educational tasks, its partnership with Reuters boosts its news capabilities. As election season approaches, Meta stresses responsible news sources, aligning with OpenAI's media partnerships for ChatGPT. These collaborations highlight a trend of tech companies selectively choosing media partners, raising transparency concerns. With OpenAI and Meta leading, AI-driven news could reshape information access, setting standards for accuracy and ethical content distribution.

Multimodal Search 🔍

Cohere Adds Image Search to Create Unified AI Search

Cohere has updated its Embed 3 search model to support image searches alongside text, allowing users to search both types of content in a unified database. This addition, aimed at businesses managing large collections of visual and textual data, helps streamline content retrieval across product images, design files, and reports without needing separate storage systems.

The upgraded system supports image embedding for PNG, JPEG, WebP, and GIF files up to 5MB, with compatibility for the model’s existing text processing. Although it only allows one image per query at present, developers can access these features via Cohere's Embed API, requiring images to be submitted as Base64-encoded data URLs.

Available on Cohere's platform, Microsoft Azure, and Amazon SageMaker, this model supports over 100 languages, pushing Cohere forward in the race for advanced multimodal search capabilities. As companies like Google and OpenAI also dive into multimodal AI, Cohere’s latest update highlights the growing demand for comprehensive, high-speed, and accurate search solutions that handle both visual and textual content.

Suggested Medium Reads✨

How MusicFX DJ is Revolutionizing AI Music Creation

Discover how Google DeepMind's MusicFX DJ is breaking new ground in AI-powered music creation, allowing anyone to generate and manipulate live music in real time. Whether you're an experienced musician or a total beginner, MusicFX DJ opens up endless creative possibilities. Read the blog⬇️  to explore this transformative tool.

Startup Success: Instant Access to AI Entrepreneur Courses!