AI News

The ENIAC, launched in 1946, was the first programmable computer, revolutionizing technology. Today, AI is evolving similarly, shifting towards edge AI, which decentralizes processing for efficiency and cost-effectiveness. This transition promises enhanced innovation but raises quality and security challenges.

Instagram has introduced Meta AI, an AI chatbot that answers questions, creates poetry, and generates images from text prompts. Currently being tested in limited capacity, it can be accessed in Instagram's search bar and Direct Messaging. It offers consistent functionality across platforms like Facebook Messenger and WhatsApp.

Elon Musk's xAI has launched Grok-1.5 Vision (Grok-1.5V), a multimodal model that processes text and visual data. It competes with leading models and outperforms them in assessments, particularly RealWorldQA. Grok-1.5V will soon be available to testers, as xAI aims to advance AI capabilities.

Google's new Infini-attention technique allows large language models (LLMs) to process infinite-length text efficiently, maintaining constant memory and computational demands. By integrating compressive memory, it enhances performance on long-context tasks, potentially transforming LLM applications.

Google's RecurrentGemma is a new open language model optimized for AI text processing on resource-constrained devices like smartphones and IoT systems. It reduces memory and processing needs while maintaining performance, enhancing edge computing capabilities and enabling real-time applications without relying heavily on cloud computing.

Amidst a tech race for larger language models (LLMs), smaller language models (SLMs) are gaining traction as performance differences narrow. SLMs offer efficiency, adaptability, lower resource needs, and enhanced privacy, positioning them as a viable alternative to LLMs for specific tasks.

Ideogram has launched an update to its text-to-image model, adding features like description-based referencing and negative prompting to enhance image creation control. With options for image generation speed and improved text rendering, the updates aim to compete with rivals like Midjourney and DALL-E.

Cohere has launched Rerank 3, a foundation model enhancing enterprise search and retrieval of complex data formats like JSON and emails, with a context length of 4,000 tokens. It offers better accuracy and cost efficiency, and integrates seamlessly with Elastic's API, emphasizing data privacy.

Apple's upcoming M4 Macs may launch later this year, following its silicon upgrade cycle. The M4 series aims for significant AI enhancements and will include updates for all Mac models by early 2025. Key features include up to 512GB memory support and integration of on-device AI processing.

OpenAI has launched GPT-4 Turbo for paid ChatGPT users, featuring enhanced math, reasoning, coding, and writing skills. It can process 300 pages of text, generating detailed responses and accommodating image prompts and text-to-speech. Its knowledge now extends to December 2023.

Rapidus, a Japanese contract chip manufacturer, has launched a U.S. subsidiary in Silicon Valley, appointing Henri Richard as general manager. The office will focus on advanced logic semiconductor design for fabless companies. Rapidus aims to enhance market entry speed and expand globally, with ongoing partnerships in 2nm technology development and strong ties to IBM. A pilot line is set to commence in 2025.

Former Google DeepMind researchers launched Udio, an AI app for quick, professional music creation. Already attracting industry interest, Udio allows users to generate songs in under a minute with automated copyright protections. It aims to democratize music production, although it faces competition and ethical challenges.