SORA Wows Creative Minds

PLUS: Orb Software Goes Open-Source, Pure Storage and Nvidia Partnership and more.

Today:

  • SORA Wows Creative Minds

  • Apple Chooses Baidu's AI

  • Tech Giants Challenge Nvidia

  • Pure Storage and Nvidia Partnership

  • Orb Software Goes Open-Source

  • Eliyan Secures $60M Boost

SORA: The STUNNING First Impressions | Artists get their hands on OpenAI's SORA and are BLOWN AWAY

OpenAI unveils third-party creations using Sora, its advanced AI model, in a series of videos by outside artists and filmmakers. Although not yet accessible to the public, these samples demonstrate Sora's potential for content creation. 

Artists like Walter Woodman and Paul Trillo showcase their work, praising Sora's capabilities. However, this promotion comes amid growing skepticism toward AI-generated content, with concerns about proper data usage and ethical implications. 

Apple to install Baidu AI on its iPhone 16 and other products in China, report says

Apple has reportedly chosen Baidu, a Chinese search engine, to provide AI technology for its upcoming iPhone 16 in China. The move aims to comply with local regulations, contrasting with Apple's use of its own AI model elsewhere. Baidu's Ernie Bot will power Apple's products in mainland China, amid Tim Cook's recent visit emphasizing commitment to the market. 

This decision follows a decline in iPhone sales amidst stiff competition from Huawei and others. Additionally, Samsung is integrating Baidu's AI into its devices, reflecting a broader trend of tech firms navigating Chinese regulations on AI services.

Nvidia’s AI chip dominance is being targeted by Google, Intel, and Arm

Major tech companies, including Intel, Google, Arm, Qualcomm, and Samsung, are collaborating to challenge Nvidia's dominance in the AI market. They aim to break Nvidia's hold by creating open-source software that allows AI developers to run their code on any machine and chip, eliminating the need for Nvidia's proprietary technology. 

The project, led by The Unified Acceleration Foundation (UXL), seeks to provide alternatives to Nvidia's CUDA platform, which currently offers superior performance but faces scarcity due to high demand. The initiative aims to broaden options for AI and high-performance computing applications while eventually supporting Nvidia's hardware and code. Additional chipmakers and cloud-computing companies are being invited to join the effort to ensure widespread adoption across different platforms.

Pure Storage, Nvidia partner to democratize AI with new infrastructure solutions

Pure Storage and Nvidia unveiled new AI infrastructure solutions at the GTC conference, aiming to streamline enterprise AI deployments. The partnership addresses challenges in fragmented data environments, offering validated reference architectures for efficient AI operations. Solutions include a Retrieval Augmented Generation (RAG) Pipeline and Nvidia OVX Server Storage certification. 

They also focus on vertical-specific AI applications like Financial RAG for the finance industry. Pure Storage expands its partner ecosystem with Run.AI and Weights & Biases to enhance AI infrastructure. Energy-efficient storage solutions are prioritized to reduce environmental impact. Pure Storage anticipates and meets AI storage needs, paving the way for future enterprise AI adoption.

WORLDCOIN FOUNDATION OPEN SOURCES CORE COMPONENTS OF THE ORB’S SOFTWARE

The Worldcoin Foundation has made a significant move by open-sourcing the core components of the Orb's software, a tool designed to verify humanness online with utmost privacy and security. This release on GitHub under an MIT/Apache 2.0 dual license enhances transparency and verifiability, crucial for ensuring trust in the Orb's image processing. 

The software addresses top-level requirements such as privacy, security, and transparency, driving its design and impacting various aspects. It leverages a robust computing stack centered around NVIDIA Jetson and incorporates Rust applications, neural networks, and secure elements to fulfill its mission effectively. 

Eliyan raises $60M for chiplet interconnects that speed up AI chips

Eliyan bags $60 million to turbocharge AI chip processing with chiplet interconnects. Samsung Catalyst Fund and Tiger Global Management lead the funding charge. The tech tackles generative AI chip development challenges amid soaring demand for AI chips. Eliyan's NuLink PHY tech ramps up performance and slashes power usage. Existing investors like Intel Capital join the funding frenzy. 

The cash injection follows a $40 million Series A round in 2022. Eliyan aims to tackle design and manufacturing hurdles for advanced AI chips with its chiplet interconnects, promising game-changing performance and efficiency. The tech also tackles memory capacity and bandwidth challenges in AI chips.

🧠RESEARCH

LLM2LLM is a method to enhance large language models (LLMs) for tasks with limited data. It iteratively augments a small dataset using a teacher LLM, improving performance without extensive manual curation. Results demonstrate significant enhancements, up to 52.6% in some cases, surpassing traditional fine-tuning methods. This approach promises more effective solutions for data-constrained applications.

Champ is a method for human image animation using a 3D parametric model within a latent diffusion framework. It employs the SMPL model to capture detailed human geometry and motion from source videos. By incorporating depth, normal, and semantic maps, alongside motion guidance, Champ generates high-quality animations with accurate pose and shape variations.

ThemeStation is a novel method for generating theme-aware 3D assets from a few exemplars. It aims to create diverse 3D models that align thematically with given exemplars. ThemeStation utilizes a two-stage framework involving concept image generation and reference-informed 3D modeling, employing a dual score distillation loss to improve quality. Results demonstrate superior performance in producing high-quality, theme-aligned 3D models, enabling controllable 3D-to-3D generation.

InternVideo2 is a leading video foundation model (ViFM) excelling in action recognition, video-text tasks, and video-centric dialogue. It employs progressive training, integrating self- and weakly-supervised learning frameworks, prioritizing spatiotemporal consistency through semantic video segmentation and captions. By scaling both data and model size, InternVideo2 achieves state-of-the-art performance across 60+ video and audio tasks, particularly in long video comprehension.

VidLA is a method for scalable video-language alignment overcoming limitations of previous approaches. VidLA employs a simple two-tower architecture, utilizing data tokens at different temporal resolutions to capture short- and long-range dependencies. Leveraging large-scale training data curated with recent LLMs, it outperforms state-of-the-art methods on retrieval benchmarks, particularly for longer videos, and competes well on classification benchmarks.

🛠️TOP TOOLS

Structify -Human quality data, super human speeds.

More Useful Things - Prompt Library.

Fructose - Python package to create a dependable, strongly-typed interface around an LLM call.

Reporfy - Design and share beautiful reports effortlessly and engage in conversation around them.

Feedhive - Create, schedule, publish, and easily manage your social media content.

📲SOCIAL MEDIA

🗞️MORE NEWS

Can you hear me now? AI-coustics to fight noisy audio with generative AI

AI-coustics, a German startup, aims to improve audio clarity using generative AI. Co-founded by audio engineer Fabian Seipel and machine learning lecturer Corvin Jaedicke, the technology enhances speech quality across various devices. With €1.9 million funding, AI-coustics employs a unique training approach to combat bias and offer clear audio solutions. TECHCRUNCH

The future of AI: Personalized systems tuned to your needs, Amazon exec says

Amazon's Vishal Sharma predicts AI's future as not one big brain but a bunch of customized helpers for everyone. Using Alexa as a springboard, Amazon envisions personal AIs doing everything from running your smart home to consulting doctors for you. Privacy issues and the need for big infrastructure could complicate things. Even so, Sharma sees it leading to an "age of abundance," despite the challenges ahead. AXIOS

FundGuard secures $100 million Series C for AI-powered investment platform

FundGuard, an Israeli fintech firm, raised $100 million in Series C funding, led by Key1 Capital and new investors. CEO Lior Yogev noted a 50% valuation increase since the last round, attributing it to client growth and market partnerships. FundGuard's AI-powered platform aids investment management and attracts top talent, primarily ex-military. CTECH

Chinese scientists create world’s most energy-efficient AI chips for mobile devices

Chinese scientists showcased two groundbreaking AI chips at a prestigious conference, boasting record energy efficiency. One chip enables offline voice control with exceptional accuracy, while the other detects epilepsy seizures, both with minimal power consumption. These innovations mark significant progress in AI chip technology for mobile devices. SCMP

Romanian state agency turns to AI to help farmers tap EU funds

A Romanian rural investment agency employs artificial intelligence (AI) to streamline EU fund applications for farmers and institutions. Robots from UiPath handle database tasks, saving 784 days of work, facilitating 5.32 billion euros in financing requests. Despite Romania's digital lag, this marks a significant step in digitalizing public services. REUTERS

AI Identifies Retinal Biomarkers for Preeclampsia

Artificial intelligence identifies retinal changes indicating preeclampsia, a leading cause of maternal death. Using the Deep Vascular Topology Network (DVT-Net), researchers analyzed 300 pregnant women's eye images, achieving an 81.3% accuracy for early preeclampsia prediction. Postpartum, increased retinal artery tortuosity suggests severe preeclampsia, underscoring AI's role in diagnosing pregnancy-related disorders. MEDSCAPE

A Psychologist Explains Why It’s Possible To Fall In Love With AI

Human-AI relationships are real, with AI becoming integral in daily life. Anthropomorphism, attributing human traits to AI, and the Triarchic Theory of Love explain why people can fall for AI. As AI mimics human behavior and offers emotional connection, users may form deep bonds, seeking qualities like empathy and stability. FORBES

Here’s Proof You Can Train an AI Model Without Slurping Copyrighted Content

OpenAI previously claimed AI models couldn't excel without copyrighted data. But recent releases challenge this, showing AI can be trained ethically. French researchers offer a large public domain dataset, while Fairly Trained certifies a model, KL3M, trained on legal documents. This shifts norms away from copyright infringement. WIRED

What'd you think of today's edition?

Login or Subscribe to participate in polls.

What are MOST interested in learning about AI?

What stories or resources will be most interesting for you to hear about?

Login or Subscribe to participate in polls.

Reply

or to participate.