Weekly AI Debrief: Apple's LLM, AI dogfights, and more!

Welcome to this week’s edition! ?? Every week (except when you start a new job, sorry people), we dive into the fascinating world of artificial intelligence, exploring groundbreaking advancements and pivotal moments shaping our future. Whether you're an AI enthusiast, a tech professional, or just curious about the digital transformations unfolding around us, you’ll find something to spark your interest.

?? Apple Releases OpenELM

Apple has recently launched OpenELM, an open-source set of large language models (LLMs) designed to operate directly on devices, rather than relying on cloud-based computing. This release includes a variety of models, tailored with different parameters to enhance on-device AI capabilities. OpenELM is notable for its enhanced accuracy and reduced dependency on extensive pre-training data compared to previous models like OLMo. Apple emphasizes that these models are a step towards greater transparency and reproducibility in AI research, aiming to support the broader open-source community by providing tools that are adaptable for use on various Apple devices (MacRumors ).

?? Eric Schmidt-backed Augment Launches

Augment, a new AI-powered coding platform, has officially launched with significant backing from Eric Schmidt and $252 million in funding. The platform is designed to enhance quality and productivity, addressing common issues in software development such as fragility and maintenance complexity. Founded by ex-Microsoft engineer Igor Ostrovsky and AI researcher Guy Gur-Ari, Augment aims to empower developers to produce high-quality software more efficiently. Despite its early stage, the company has already attracted interest from hundreds of developers and aims to differentiate itself in the competitive AI coding assistant market, which includes established players like GitHub Copilot and newcomers like CodeWhisperer and Gemini Code Assist (TechCrunch ).

??? The Ray-Ban Meta Smart Glasses Update

The Ray-Ban Meta smart glasses have been upgraded with new multimodal AI capabilities, enhancing their functionality significantly. These smart glasses, now equipped with Meta AI, offer users a variety of interactive features, such as object recognition and real-time translation. This multimodal AI allows the glasses to process audio and visual input simultaneously, providing users with information about their environment directly through the glasses. For example, users can get descriptions and relevant data about objects they see, or receive translations for foreign language texts in their view (Facebook ) (Ray-Ban ).

?? OpenAI's New Enterprise AI Features

OpenAI has recently introduced ChatGPT Enterprise, an advanced version of their popular AI model designed for business use. This new enterprise-level service enhances privacy and security for corporate data, provides unlimited access to high-speed GPT-4 capabilities, and offers extensive customization options to meet diverse organizational needs. Key features include a sophisticated admin console for user management, single sign-on, domain verification, and detailed usage analytics. ChatGPT Enterprise also expands the context window to handle much larger inputs and improves response times, making it ideal for complex data analysis and high-volume interactions (OpenAI ).

?? Microsoft Releases Phi-3

Microsoft has recently launched its Phi-3 series of compact language models, aimed at delivering powerful AI capabilities with reduced computational demands. The Phi-3 series includes several models like the Phi-3-mini, Phi-3-small, and Phi-3-medium, which vary in complexity and size, making them suitable for different applications from mobile devices to enterprise solutions (Microsoft ).

??? US Air Force AI Dogfight

The U.S. Air Force has confirmed its first successful dogfight involving an AI-controlled fighter jet, specifically the X-62A VISTA, against a human-piloted aircraft. This milestone event, conducted under the Air Combat Evolution (ACE) program by DARPA, marks a major advancement in the use of AI in aerial combat. The test took place at Edwards Air Force Base in California, highlighting the AI's capability to handle complex flight maneuvers against a human opponent. This test represents a breakthrough in military aviation, demonstrating potential future applications of AI in enhancing combat strategy (Fox Business ).

?? Adobe's New Firefly Model

Adobe has introduced updates to its Firefly model, enhancing its capabilities for AI-driven visual content generation across its Creative Cloud and Adobe Express platforms. The new updates include the Firefly Image 2 Model, which offers advanced image generation with improved control and quality. Additionally, the Firefly Vector Model, a pioneering AI model for vector graphics, is now integrated into Adobe Illustrator, enabling the creation of high-quality vector graphics from text prompts (Adobe News ).

?? Brave Search Adopts AI

Brave Search introduced a new AI-powered feature called "Answer with AI," designed to enhance user experience by providing direct answers to queries within the search results. This feature uses a combination of large language models and Brave's index to generate synthesized responses from various sources, ensuring privacy and reducing reliance on external search engines like Google and Bing (TechCrunch ).

?? Hollywood Begins AI Cloning

Hollywood is venturing into AI cloning, with the technology now able to scan the bodies, faces, and voices of actors to create lifelike digital doubles. This has sparked considerable debate and concern within the industry. Major talent agencies and studios are exploring how these AI clones can be utilized for roles in movies and advertisements, potentially reducing costs and logistical challenges associated with scheduling and physical presence. However, this has led to pushback from actors and their unions, who are worried about the implications for their careers and the authenticity of their performances. Issues such as consent, compensation, and the ethical use of their digital likenesses are at the forefront of ongoing discussions and negotiations (The Rundown AI ).

?? Meta's Next-gen Llama 3

Meta has officially released its third-generation large language model, Llama 3, which showcases major enhancements over its predecessors. This model includes improved training mechanisms and has been fine-tuned for various use cases, from conversational AI to content creation, ensuring better performance across multiple tasks. Llama 3 is built with a new tokenizer and enhanced context windows that improve its understanding and generation capabilities, especially in handling complex and nuanced language tasks. Additionally, the model's performance benefits from a substantial increase in training data, which Meta has curated to ensure high quality and relevance (Meta AI ).

???? Chinese AI Model Surpasses GPT-4 Turbo

Recently, China's Baidu and iFlytek have made claims that their respective AI models, Ernie 4 and Spark v3.5, surpass OpenAI's GPT-4 Turbo in several aspects. Baidu's Ernie 4 reportedly excels in tasks involving the Chinese language, specifically in generating texts in the style of Imperial Chinese poetry. This specialization highlights its capabilities in nuanced language tasks, which are culturally and contextually rich (South China Morning Post ).

Thanks for joining us for another exciting issue of the Weekly AI Debrief! We hope you enjoyed the latest and greatest in AI innovation. Stay curious, stay inspired, and remember—the future isn’t just something we enter, it’s something we create. Until next week, keep pushing the boundaries of what's possible! ??