Today:

• Meta Unveils Llama 3

• Meta Integrates Smart Assistants Everywhere

• Microsoft's VASA-1 Brings Photos to Life with AI

• Oracle's $8 Billion Investment in Japan

• Cisco AI-Powered HyperShield

• YouTube's AI-Generated Deepfake Porn

Meta's LLAMA 3 SHOCKS the Industry | OpenAI Killer? Better than GPT-4, Claude 3 and Gemini Pro

Meta Llama 3, surpassing expectations, releases its models with 8 billion and 70 billion parameters, boasting top-notch performance. More models, including a 400 billion-parameter one, are in the works. Mark Zuckerberg emphasizes the importance of open-sourcing AI to prevent concentration of power, highlighting risks of centralized AI control. 

The upcoming 400 billion-parameter Llama 3 model could rival top proprietary models, fostering a more balanced AI landscape. Synthetic data usage for training raises questions about model scalability, with constraints likely preventing runaway AI growth. Open-source ethos drives Meta's push, challenging industry norms. 

WATCH THE VIDEO ON YOUTUBE

Meta, in Its Biggest A.I. Push, Places Smart Assistants Across Its Apps

Meta is rolling out new versions of its AI-powered smart assistant, Meta A.I., across its apps like Instagram, WhatsApp, Messenger, and Facebook. These assistants, powered by the latest LLaMA 3 AI model, will offer users assistance with tasks and information retrieval. 

Meta's move marks its largest deployment of AI technology yet, aiming to integrate AI products into the most visible parts of its apps. The company's open-source approach distinguishes it, allowing others to utilize and improve the underlying technology. The rollout reflects Meta's commitment to enhancing user experiences and embracing the potential of AI across its platforms.

THE NEW YORK TIMES

Microsoft shows off VASA-1, an AI framework that makes human headshots talk, sing

Microsoft unveiled VASA-1, an AI framework that animates static headshots, making them talk and sing realistically. The innovation requires minimal input—just a headshot and audio file—to generate lifelike videos with synchronized lip movements and expressions. Despite its potential for deepfake misuse, Microsoft stresses its research-only status, with no plans for commercialization. 

VASA-1 presents a balancing act, offering positive applications like advertising while acknowledging risks. Microsoft aims for responsible AI integration, envisioning applications in education and accessibility. However, challenges remain, including smoothness in movement and potential misuse, prompting cautious advancement.

VENTUREBEAT

Oracle to invest over $8 billion in Japan in cloud computing, AI

Oracle announced a significant investment of over $8 billion in Japan over the next decade to meet the growing demand for cloud computing and AI infrastructure. The plan includes expanding Oracle Cloud Infrastructure (OCI) services across Japan and increasing operational and support teams in the country. 

This move aligns with Oracle's strategy to capitalize on the expanding market for cloud services and underscores its commitment to the Japanese market. The investment aims to enhance Oracle's presence in Japan and strengthen its position in the competitive cloud computing and AI sectors.

REUTERS

Cisco debuts new AI-focused security system after $28 billion deal to buy Splunk

Cisco introduced HyperShield, a new security system leveraging AI to safeguard applications, devices, and data across various IT environments. This move follows Cisco's $28 billion acquisition of Splunk and aims to address evolving cybersecurity needs. HyperShield employs autonomous segmentation to enhance visibility and security, with features like self-qualifying upgrades for seamless system maintenance. 

Designed with the rise of digital AI assistants in mind, it responds to the growing complexity of modern applications and technologies. The introduction of HyperShield underscores Cisco's commitment to cybersecurity amidst shifting market dynamics.

CNBC

Google Hosted Over 100 YouTube Videos Promoting AI Deepfake Porn

Google faced scrutiny after Forbes discovered over 100 YouTube videos promoting AI deepfake porn, leading to their removal. These videos advertised apps that could undress people in images, with some tutorials linked to instances of bullying and exploitation. Despite Google's efforts to remove offending content, issues persisted on other platforms like Android apps and Google Ads. 

The National Center on Sexual Exploitation highlighted Google's role in profiting from these apps and urged for responsible policies. Concerns grow as AI-generated deepfake porn, including of children, becomes more prevalent, raising alarming implications for victims' privacy and safety.

FORBES

🧠RESEARCH

VASA-1: Lifelike Audio-Driven Talking Faces Generated in Real Time

VASA-1 by Microsoft Research Asia introduces lifelike talking faces from a single image and audio clip. VASA-1 excels in lip syncing, facial expressions, and head movements, enhancing authenticity and liveliness. Innovations include diffusion-based facial dynamics and an expressive face latent space. Outperforming predecessors, it offers high-quality, real-time avatars for human-like interactions. 

A Survey on Retrieval-Augmented Text Generation for Large Language Models

The survey explores Retrieval-Augmented Generation (RAG), merging retrieval methods with deep learning to enhance large language models (LLMs). It categorizes RAG into pre-retrieval, retrieval, post-retrieval, and generation phases, detailing its evolution and evaluation methods. The paper aims to consolidate RAG research, elucidate its technology, and expand LLM applications.

The Landscape of Emerging AI Agent Architectures for Reasoning, Planning, and Tool Calling: A Survey

The survey explores recent advances in AI agent architectures, emphasizing their capacity for complex tasks like reasoning, planning, and tool execution. It aims to convey current capabilities and limitations, share insights from practical use, and suggest considerations for future design. The paper covers single-agent and multi-agent architectures, highlighting design patterns, leadership roles, communication styles, and planning phases for robust AI systems.

How faithful are RAG models? Quantifying the tug-of-war between RAG and LLMs' internal prior

The study examines the reliability of Retrieval Augmented Generation (RAG) in correcting errors made by Large Language Models (LLMs). While providing correct retrieved content usually fixes mistakes, when the retrieved information is incorrect, LLMs may still recapitulate errors. The research highlights a tension between a model's prior knowledge and external information.

Reducing hallucination in structured outputs via Retrieval-Augmented Generation

The paper addresses the challenge of hallucination in Generative AI (GenAI), particularly in Large Language Models (LLMs). By employing Retrieval-Augmented Generation (RAG), the proposed system significantly reduces hallucination in structured outputs, enhancing LLM generalization across domains. Additionally, a well-trained retriever reduces the size of accompanying LLMs, making deployments more resource-efficient.

🛠️TOP TOOLS

Package Design - Generate product packaging designs tailored to your theme & style

Ocular AI - Search, visualize, and take action on your work and engineering tools and data on one unified platform

Assembly - Designed for recognition. Optimized for engagement.

Maihem - AI agents that continuously test your AI applications

Rely - Promote standards, developer autonomy and visibility across your software delivery lifecycle.

📲SOCIAL MEDIA

🗞️MORE NEWS

To Unlock AI Spending, Microsoft, OpenAI and Google Prep ‘Agents’

Microsoft, OpenAI, and Google are enhancing conversational AI capabilities to incentivize businesses to invest in AI technology. Microsoft plans to unveil software at its Build conference, powered by OpenAI, enabling automation of tasks like invoice creation and code rewriting. This move aims to make AI more indispensable for businesses, despite lingering spending hesitations. THE INFORMATION

Meta is adding real-time AI image generation to WhatsApp

Meta is introducing real-time AI image generation in WhatsApp for US users. When you type, the image changes based on your description. The Meta Llama 3 model now creates sharper images and can animate them. This feature extends to Meta's apps, including WhatsApp, Instagram, Facebook, and Messenger. THE VERGE

Dataminr debuts ReGenAI pairing predictive with generative AI for real time information

Dataminr introduces ReGenAI, blending predictive and generative AI to dynamically update event alerts in real time. Unlike traditional prompt-based systems, ReGenAI autonomously synthesizes and reformulates information from diverse data sources, providing organizations like the UN and Airbus with real-time insights and recommendations. CEO Ted Bailey emphasizes its evolution beyond mere prompts, leveraging internal large language models for customized, third-party-independent functionality. VENTUREBEAT

Slack AI is now available to all paid users. Here's what these features can do for your team

Slack's AI add-on, now available to paid users, enhances team communication and productivity. Features include daily summaries, AI-powered search, and conversation summaries. Slack assures users of data security, stating customer data isn't used for training AI models. Priced at $10/user/month, it supports English, Spanish, and Japanese, with more languages planned. ZDNET

Blockade Labs improves quality for AI-generated 3D art for 360-degree apps

Blockade Labs unveils Skybox AI update, enhancing 360-degree art creation with 8K quality, ideal for Apple Vision Pro. CEO deCourcelle highlights its applications across sectors and partnerships with major companies. The Model 3 update improves resolution by 77%, while features like remixing and AI-driven sketch recognition streamline the creative process. VENTUREBEAT

US Air Force confirms first successful AI dogfight

The US Air Force tested AI against a human pilot in a dogfight, marking a milestone in AI aviation. DARPA, in charge of the experiment, reports success with an AI-controlled jet facing off against a human-piloted F-16. The test, part of the ACE program, signals advancements in autonomous flight. THE VERGE

NIH researchers develop AI tool with potential to more precisely match cancer drugs to patients

NIH researchers developed an AI tool using single-cell RNA sequencing to predict cancer drug responses. Unlike bulk sequencing, this approach targets individual cells, potentially leading to more effective treatments. The AI accurately predicted drug responses in cell-line data and patient studies. While reliant on costly data, the tool, PERCEPTION, shows promise for personalized cancer treatment. NIH

ChatGPT 4 could be used to triage eye problems, say Cambridge academics

Cambridge academics suggest using ChatGPT 4 to triage patients with eye issues, as it performed on par with experts in a study. The AI scored 69% in a mock exam, outperforming previous versions and other models. While not a replacement for doctors, it could assist in determining urgent cases and providing advice to non-specialists, potentially reducing wait times for eye care in the UK. BBC

WHO’s New AI-Powered Chatbot Is Giving Wrong Medical Answers

The World Health Organization introduces SARAH, a virtual health worker powered by AI, aimed at providing basic health information and filling staffing gaps. While available in multiple languages and responsive to users' facial expressions, SARAH's reliance on outdated data sometimes leads to inaccurate responses, prompting a need for improvement. BLOOMBERG

AI optimists crowd out doubters at TED

At TED2024, tech leaders like Microsoft's Mustafa Suleyman envision AI as a new digital species, promising transformative breakthroughs in healthcare and education. However, concerns over AI's limitations and potential societal harms persist, prompting calls for greater transparency, auditing, and collaboration within the industry. While AI dominates discussions, cities like Boston and Northwest Arkansas vie for prominence in the AI job market, highlighting the broader impact of AI beyond the TED stage. AXIOS

How AI Tools Can Make More Money For Small Businesses

AI tools are reshaping small businesses, offering efficiency boosts and cost savings. They automate tasks, enhance customer service, optimize inventory, refine marketing, streamline finances, and foster innovation. By starting small and scaling gradually, businesses can harness AI's potential for success in today's competitive market. FORBES