AI Updates September 2024 - Monthly Overview

Mira Murati Resigns Amid Leadership Overhaul and Structural Changes at OpenAI

Mira Murati, OpenAI's Chief Technology Officer, along with two other senior executives, has resigned amidst significant structural changes within the organization. OpenAI is transitioning from its nonprofit roots towards becoming a B Corporation, signaling a shift towards a traditional corporate structure with increased investor focus. While the leadership changes, including new appointments and Murati's transitional role, mark a major overhaul, CEO Sam Altman remains optimistic about OpenAI’s future, despite growing competition in the AI space.

Meta AI Unveils Llama 3.2

Meta AI's Llama 3.2, released on September 25, 2024, introduces advanced large language models optimized for edge computing and multimodal processing. It includes lightweight text-only models for mobile devices and multimodal models that integrate vision and language for tasks like image captioning and document understanding. With extended context lengths of up to 128K tokens, efficient on-device processing, and improved performance, Llama 3.2 offers versatile solutions for AI development. Meta continues its commitment to open-source innovation, providing powerful tools for developers while partnering with major cloud and hardware providers for widespread adoption.

Snapchat Unveiled "Snap AI Video"

Snapchat has introduced Snap AI Video, an AI-driven tool that allows creators to generate videos from text prompts, announced in beta at the annual Snap Partner Summit. With plans to expand access, the tool simplifies video production for creators, potentially revolutionizing social media content creation. Snap AI Video also prioritizes safety with rigorous testing and watermarked AI-generated content. This move positions Snapchat ahead of competitors like TikTok and Instagram in AI integration, attracting interest from creators, brands, and advertisers while setting the stage for a new era of AI-driven content on social media.

Adobe New Firefly video model

Adobe has announced its new Firefly video model, a groundbreaking AI tool set to revolutionize video editing. Building on the success of its generative AI model for images, Firefly now introduces features like text-to-video and image-to-video generation, along with advanced tools like generative extend, which seamlessly fills gaps in footage. The tool simplifies and speeds up video production, making it easier for creators to generate high-quality, engaging content. Firefly is also ethically designed, using only legally permissible content, and will roll out in beta later this year, offering a transformative tool for video creators.

Arcee-SuperNova: Flagship 70B LLM for Enterprises

Arcee AI has launched SuperNova, a new AI model designed to rival large language models like OpenAI's. Available in two versions—SuperNova 70B and SuperNova-Lite 8B—SuperNova excels at instruction-following, mathematical reasoning, and data privacy. Using innovative techniques like model distillation and merging, SuperNova offers superior performance while allowing enterprises full ownership and customization of the model, ensuring data security and reducing costs by up to 75%. With flexible deployment options, SuperNova is a powerful alternative for businesses seeking advanced AI solutions without compromising control or privacy.

OpenAI Academy

OpenAI Academy aims to empower developers in low- and middle-income countries by offering $1 million in API credits, training, and support. The initiative seeks to boost innovation in sectors like healthcare and education. While some critics raise concerns, it provides valuable AI tools to help local developers drive change.

OpenAI o1-preview

OpenAI has launched the o1 series of AI models, designed for advanced problem-solving in fields like science, coding, and mathematics. These models excel in reasoning, achieving high performance in tasks such as math competitions and coding challenges. With enhanced safety protocols, the o1 models are now available to select users through ChatGPT and API access.

OpenAI 'GPT-Next'

OpenAI Japan has announced GPT-Next, a groundbreaking AI model expected to be 100 times more powerful than GPT-4, launching by the end of 2024. With smarter architecture and enhanced multimodal capabilities, GPT-Next will offer faster, more efficient problem-solving across fields like science and programming. This advancement marks a significant leap in AI technology, sparking excitement and anticipation in the tech community.

Mercado Libre ‘Verdi’

Mercado Libre has revolutionized its operations using AI, notably through its Verdi platform, which automates complex tasks like customer service mediation and fraud detection. Verdi, powered by GPT-4 and GPT-3.5, enables scalability and efficiency, handling tasks previously requiring human intervention. As Mercado Libre expands Verdi's applications, the platform is poised to enhance areas like supply chain management, predictive analytics, and financial services.

NVIDIA’s Llama-3.1-Nemotron-51B

NVIDIA's Llama-3.1-Nemotron-51B, derived from Meta’s Llama-3.1-70B, is optimized to run efficiently on a single H100 GPU, offering performance similar to much larger models. By using Neural Architecture Search and knowledge distillation, the model delivers 2.2x faster inference and 4x larger workload capacity, reducing costs and enhancing accessibility. This breakthrough makes high-performance AI more practical for businesses and real-world applications.

🆓🖥️🔍🌐

Microsoft Launched "GRIN-MoE"

GRIN-MoE is an AI model that uses Gradient-Informed Routing and Mixture-of-Experts architecture to activate only task-specific parameters, reducing computational costs while maintaining high performance. It excels in specialized tasks like coding and mathematical problem-solving, outperforming larger models with fewer active parameters. While highly efficient, GRIN-MoE struggles with broader natural language tasks, offering room for future improvement.

🕸️🤖💻📊

AI-RAN and NVIDIA AI Aerial Revolutionizing Wireless Networks

NVIDIA AI Aerial is an AI-driven platform designed to optimize wireless networks, enabling telecom providers to support advanced technologies like 5G, autonomous systems, and generative AI. By integrating AI with radio access networks (AI-RAN), it enhances performance, reduces costs, and creates new revenue opportunities. This platform, supported by key industry collaborations, is set to revolutionize telecommunications.

💳🤖💼🌐

NVIDIA Medusa Boosts AI

NVIDIA's Medusa algorithm optimizes the Llama 3.1 model, delivering a 1.9x performance boost by accelerating token generation and reducing latency. By using speculative decoding and TensorRT-LLM for efficient multi-GPU processing, Medusa enhances real-time AI applications, making large language models faster and more accessible.

China makes AI breakthrough, reportedly trains generative AI model across multiple data centers and GPU architectures

Apple Reportedly Exits OpenAI’s $6.5 Billion Funding Round

73% of AI pros are looking to change jobs over the next year

AMD unveils its first small language model, AMD-135M — AI performance enhanced by speculative decoding

OpenAI funding and restructuring plans renew pressure on AI’s top startup

The world's first AI art museum will be a big test for the technology

👥 Connect & Feedback!

👉 Join Us:

📧 Advertise In Weekly AI News:

📧 Contact directly at [email protected]

😍 Share with your friends!