Read time: 10 minutes | Sponsor this newsletter

Welcome to the weekly AI report

In this week's AI news:

• Gemini: Google's Revolutionary AI Model

• xAI’s Grok is rolling out

• Relightable VR Avatars based on Gaussian Splatting

• Meta Unveils 20+ AI Enhancements Across Social Platforms

• Meta Launches Purple Llama for Enhanced AI Trust and Safety

• OpenAI's $51 Million AI Chip Deal with Startup Rain AI Raises Conflict of Interest Concerns

• IBM and Meta Co-Launch AI Alliance: A Global Collaboration for Open, Safe, Responsible AI

• Microsoft Copilot Celebrates a Year of AI Milestones and Unveils Upcoming Innovations

• DeepMind Develops AI Agents Capable of Cultural Transmission Through Few-Shot Imitation

• Runway and Getty Images Collaborate to Revolutionize Enterprise Video Creation with AI

Gemini: Google's Revolutionary AI Model

Google has finally joined the race, at least in some form, as Gemini Pro is now available. It is slightly better than GPT 3.5 and can currently be tested in Bard. However, we will have to wait until the beginning of next year for Gemini Ultra.

According to announcements and their test results, Gemini Ultra has achieved a higher score than OpenAI's GPT-4 in all tests, making it the best LLM so far.

What sets Gemini apart from GPT-4 is that it is initially trained on multimodal datasets. This means that the base model naturally accepts input in the form of text, images, audio, and even video although it converts video into image sequences, but users won’t be aware of it, thanks to Google's incredible computing power).

Key Points:

• Gemini, developed by Google DeepMind, represents a major advancement in AI, offering state-of-the-art performance across numerous benchmarks.

• This multimodal model excels in understanding and integrating different types of information like text, code, audio, image, and video.

• Gemini is available in three versions: Ultra for complex tasks, Pro for a range of tasks, and Nano for on-device applications.

• The model demonstrates superior performance in various domains, notably outperforming human experts in multitask language understanding.

• Gemini's creation involved Google's advanced infrastructure and TPUs, ensuring reliability, scalability, and efficiency.

Significance:

If everything goes as Google says (and there are well-founded doubts that the demo they showed was not a live demo, and they already have a history of not meeting expectations), this will be a strong blow for OpenAI and Microsoft, who will have to release something new to the market. very soon to bring back the users. All in all, the AI race is heating up, which will ultimately be good for users.

You can read more about Gemini here.

xAI’s Grok is rolling out

• Grok has been released for Premium+ users of Twitter/X.

• Grok will initially be available to users in the US, but if you are not in the US, you can use a VPN to access this tool.

• Grok has a regular mode, but it also has a Fun mode where Grok's sense of humor shines.

• The advantage of Grok over all other LLMs is that it has real-time access to all information from Twitter/X.

Relightable VR Avatars based on Gaussian Splatting

Relightable Gaussian Codec Avatars introduce a groundbreaking method for creating and animating ultra-realistic, relightable head avatars, achieving unparalleled detail and real-time performance in virtual reality.

Key Points

• Advanced Geometry and Lighting: Utilizes 3D Gaussians to capture intricate details like hair strands and pores, paired with a novel relightable appearance model for realistic lighting effects.

• Efficient Relighting and Animation: Offers real-time relighting under various conditions and explicit control over expressions, gaze, and lighting, enhancing interactive VR experiences.

• Superior Performance: Outperforms existing methods in fidelity without sacrificing real-time capabilities, demonstrated on consumer VR headsets.

• Versatile Application: Suitable for interactive VR rendering and video-driven animation, featuring disentangled control over expression, gaze, viewpoint, and lighting.

• Innovative Technique: Integrates learnable radiance transfer and spherical harmonics for all-frequency reflections, ensuring dynamic, high-quality rendering.

Significance

This development marks a significant leap in virtual reality technology, offering an unprecedented level of realism and interactivity for VR avatars, crucial for the immersive experience in virtual environments.

Meta Unveils 20+ AI Enhancements Across Social Platforms

Meta is revolutionizing the user experience across its social media platforms by testing over 20 new generative AI features. These enhancements, spanning Facebook, Instagram, Messenger, and WhatsApp, aim to improve functionalities in areas like search, social discovery, ads, and business messaging. A notable addition is the introduction of invisible watermarking for AI-generated images, ensuring transparency and traceability.

Key Points:

• Generative AI Integration: Meta is embedding generative AI across its key platforms, enhancing various user experiences.

• Meta AI Features: The upgrade includes a more capable Meta AI, offering detailed responses and accurate search summaries. Additionally, "imagine with Meta AI" has been introduced for creative image generation.

• Social Interaction with "Reimagine": This feature in Meta AI allows users to interactively modify AI-generated images in a social context.

• Enhancing Facebook User Experience: AI is being leveraged to improve content creation, foster group interactions, and optimize Marketplace operations on Facebook.

• Invisible Watermarking: An upcoming feature for AI-generated images, promoting transparency and traceability.

Significance:

The substantial expansion of AI capabilities across Meta's platforms highlights the company's dedication to integrating advanced AI technology. This move not only enhances user interactions and content creation but also sets a new benchmark in the social media domain. Meta's commitment to advancing AI applications signifies a transformative phase for digital experiences, underscoring the profound impact of AI in shaping the future of social media interactions.

Meta Launches Purple Llama for Enhanced AI Trust and Safety

Meta introduces Purple Llama, a comprehensive project offering tools and evaluations for developers to build AI models responsibly, focusing on cybersecurity and input/output safeguards.

Key Points:

• Purple Llama is a new initiative by Meta to provide trust and safety tools for responsible AI development.

• The project includes tools for cybersecurity and input/output safeguards to address risks associated with Large Language Models (LLMs).

• Purple Llama's cybersecurity component offers industry-wide safety evaluations and tools to mitigate LLM cybersecurity risks.

• The input/output safeguard, Llama Guard, aims to prevent the generation of risky or inappropriate content.

• Purple Llama emphasizes an open ecosystem approach, collaborating with partners like AWS, Google Cloud, Nvidia, and others.

Significance:

Purple Llama represents a significant step in ensuring safer and more responsible AI development, aligning with Meta's commitment to open, collaborative AI research and development, and addressing growing concerns around AI ethics and security.

OpenAI's $51 Million AI Chip Deal with Startup Rain AI Raises Conflict of Interest Concerns

OpenAI, under CEO Sam Altman, signed a letter of intent to purchase $51 million in AI chips from startup Rain AI, a company in which Altman has personally invested.

Key Points:

• OpenAI agreed to spend $51 million on neuromorphic processing units (NPUs) from Rain AI, a startup backed by CEO Sam Altman.

• Rain AI, located near OpenAI's headquarters, is developing NPUs that mimic the human brain's features.

• The deal highlights potential conflicts of interest due to Altman's personal investment in Rain AI.

• OpenAI has not moved forward with the deal yet, citing only initial discussions and expressing openness to future talks.

• Rain AI's technology promises significant advancements in computing power and energy efficiency for AI development, but faces challenges due to a recent reshuffle in leadership and investment concerns.

Significance:

This agreement underscores the complexities and potential conflicts in the rapidly evolving AI industry, where personal investments of company leaders can intertwine with corporate decisions, particularly in the high-stakes area of AI chip development.

IBM and Meta Co-Launch AI Alliance: A Global Collaboration for Open, Safe, Responsible AI

IBM and Meta, in collaboration with over 50 global organizations, have launched the AI Alliance, aiming to foster open innovation and responsible AI development.

Key Points:

• The AI Alliance, co-launched by IBM and Meta, includes leading technology developers, researchers, and adopters from diverse sectors.

• Over 50 founding members and collaborators are part of the Alliance, including AMD, CERN, Google Cloud, Intel, and many prestigious universities.

• The Alliance focuses on promoting open, transparent AI innovation prioritizing safety, diversity, and economic benefits.

• Objectives include developing AI benchmarks and standards, advancing open foundation models, supporting AI hardware ecosystems, global AI skill-building, and informing public discourse on AI.

• The AI Alliance will operate through member-driven working groups, a governing board, and a technical oversight committee to oversee project areas and establish standards.

Significance:

The formation of the AI Alliance marks a significant step in uniting global efforts to ensure that AI development is open, responsible, and beneficial to society, reflecting the diverse needs of the international community and advancing AI technology in ethical and sustainable ways.

Microsoft Copilot Celebrates a Year of AI Milestones and Unveils Upcoming Innovations

Microsoft marks Copilot's first anniversary with significant advancements and new features, showcasing the impact of AI in enhancing daily digital tasks and productivity.

Key Points:

• Microsoft Copilot has integrated AI into daily tasks, influencing how people search, shop, code, and create content.

• New features include GPT-4 Turbo for more complex tasks, an updated DALL-E 3 model for higher-quality images, and Inline Compose in Microsoft Edge.

• Copilot is also introducing Multi-Modal with Search Grounding and a Code Interpreter for complex calculations and coding tasks.

• Deep Search in Bing, using GPT-4, will optimize search results for complex topics.

• Copilot is now accessible to everyone on any device through copilot.microsoft.com.

Significance:

Copilot's new features and integration across Microsoft's suite of products demonstrate a significant leap in AI capabilities, enhancing productivity and creativity, and solidifying Microsoft's position as a leader in AI-driven technology solutions.

DeepMind Develops AI Agents Capable of Cultural Transmission Through Few-Shot Imitation

Artificially intelligent agents can now acquire and use information from each other in real-time, demonstrating high fidelity and recall, akin to human cultural transmission, through a method called few-shot imitation.

Key Points

• Cultural Transmission in AI: The study introduces a method enabling AI agents to imitate human behavior in real-time within novel contexts, without needing pre-collected human data.

• Simple Ingredients for Complex Learning: A surprisingly simple set of ingredients was identified as sufficient for facilitating this cultural transmission, including memory, presence of an expert, and attentional bias.

• Real-Time Imitation and Recall: The agents demonstrated the ability to imitate human co-players in a 3D simulation environment, recalling and applying learned behaviors even after the human expert's departure.

• Methodology and Environment: The research used deep reinforcement learning and a new 3D simulation environment called GoalCycle3D, offering diverse task variants for the agents.

• Implications for AI Development: This approach paves the way for cultural evolution to play a significant role in developing artificial general intelligence, as it enables efficient exploration and adaptation in AI agents.

Significance

This breakthrough in AI learning mimics human-like cultural transmission, marking a significant step towards more adaptive and intelligent AI systems capable of learning directly from human behavior in diverse contexts.

Runway and Getty Images Collaborate to Revolutionize Enterprise Video Creation with AI

Runway is partnering with Getty Images to launch a new AI-powered video model for enterprise customers, aimed at revolutionizing high-quality, customized content creation.

Key Points:

• Runway and Getty Images are launching the Runway <> Getty Images Model (RGM), an AI video model for enterprises.

• RGM allows companies to build custom models for video content generation, using their proprietary datasets.

• This partnership targets various industries, including Hollywood, advertising, and media, enhancing creative capabilities and offering new channels for video creation.

• The collaboration emphasizes a blend of human creativity with AI technology, aiming to unlock new commercial uses and video products.

• RGM will be commercially available soon, offering a significant tool for companies with unique, proprietary datasets.

Significance:

This partnership marks a significant advancement in AI-assisted video production, catering to the growing demand for tailored, professional content in various industries.

Thank you for reading!

That is all for this week's Weekly AI report. If you liked this one, be sure to follow me on X and LinkedIn. Until the next Friday!

Take the purple pill and stay in wonderland