Today:

Gemini Pro Dominates LLM Ranks | New January Model Stealth Release?

Google's AI model, Bard, has made a huge leap in the Chatbot Arena's leaderboard, surpassing GPT-4 and landing in second place. This arena is a testing ground where different language models are anonymously pitted against each other, and users globally vote for the best responses to their prompts. 

In other Google news, CEO Sundar Pichai's leaked memo reveals big AI goals for 2024, but it's mixed with corporate jargon and hints at layoffs. Google's workplace culture seems to be shifting, with reports of employee cynicism and disillusionment.

WATCH THE VIDEO ON YOUTUBE

Elon Musk’s AI start-up seeks to raise $6bn from investors to challenge OpenAI

Elon Musk's AI start-up, xAI, is in talks to raise up to $6bn from global investors to challenge Microsoft-backed OpenAI. Musk has courted wealthy individuals and investors around the world, including in Hong Kong, and hopes to raise the funds at a proposed valuation of $20bn.

The fundraising efforts involve targeting family offices in Hong Kong, sovereign wealth funds in the Middle East, and investors in Japan and South Korea. The scale of the fundraising reflects the high costs of developing generative AI models, which require significant computing power, vast amounts of data, and cutting-edge chips.

FINANCIAL TIMES

OpenAI and Other Tech Giants Will Have to Warn the US Government When They Start New AI Projects

The US government will require tech companies like OpenAI, Google, and Amazon to provide advance warning of new AI projects involving large language models. The government will have access to key information about sensitive AI projects, including safety testing.

The new requirement is part of a sweeping executive order issued in October 2021 and is expected to take effect soon.

WIRED

Samsung’s new phones replace Google AI with Baidu in China

Samsung's latest Galaxy S24 series in China is ditching Google's AI for Baidu's Ernie. This big change means Samsung's phones in China will use Baidu's tech for things like translation and making sense of long texts. Google, big in the West, isn't part of this Chinese launch. This swap is likely because Google's not really a thing in China. 

Samsung, once a smartphone king in China with a 20% market share, has seen its popularity drop big time. Now, it's barely hanging on with just 1% market share. Teaming up with local big shots like Baidu is Samsung's play to win back Chinese customers.

THE VERGE

🧠RESEARCH

Inflation with Diffusion: Efficient Temporal Adaptation for Text-to-Video Super-Resolution

"Inflation with Diffusion: Efficient Temporal Adaptation for Text-to-Video Super-Resolution" proposes a new diffusion-based approach for enhancing video quality using text. It inflates the text-to-image super-resolution model for video generation, ensuring temporal coherence across frames. The method is evaluated on the Shutterstock video dataset, showing impressive visual quality and temporal consistency in video super-resolution. It represents a significant step in video enhancement technology.

modelscope/scepter_studio

Scepter Studio by Alibaba is an AI-powered space for generating and editing images using text descriptions. It's designed for creative applications, allowing users to transform their ideas into visual forms easily. The tool leverages advanced AI technology, making it a valuable asset for artists, designers, or anyone looking to explore the intersection of art and AI. 

OrionStarAI/Orion-14B-App-Demo-CN

The Orion-14B-App-Demo-CN by OrionStarAI, is a demonstration of an application utilizing the Orion-14B language model. This demo showcases the capabilities of the model in understanding and responding to user inputs. It's a valuable resource for those interested in exploring the functionalities of advanced AI language models in real-world applications. 

VMamba: Visual State Space Model

"VMamba" is a new architecture for visual representation learning, merging the best of Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs). It uses a state space model approach for linear complexity, maintaining global receptive fields. The Cross-Scan Module (CSM) in VMamba efficiently processes images, converting them into sequences for better analysis. This model shines in various visual tasks and outperforms others, especially with high-resolution images.

Fast Registration of Photorealistic Avatars for VR Facial Animation

The paper presents a new system for creating realistic, virtual reality (VR) facial animations using a headset-mounted camera. It overcomes the challenges of domain differences between avatars and camera images with two innovative modules: an iterative refinement module for in-domain inputs and a style transfer module guided by avatar expressions and head pose. This approach ensures high-quality, efficient VR facial animation without needing costly offline registration, significantly outperforming existing methods.

EmerDiff: Emerging Pixel-level Semantic Knowledge in Diffusion Models

"EmerDiff: Emerging Pixel-level Semantic Knowledge in Diffusion Models," explores leveraging pre-trained diffusion models like Stable Diffusion (SD) to generate detailed image segmentation maps without additional training. It highlights the challenge of extracting pixel-level semantic relations from lower-dimensional feature maps in diffusion models and offers a solution by identifying semantic correspondences between image pixels and these feature maps. The resulting segmentation maps show high accuracy and detailed capture of image parts, proving the existence of precise pixel-level semantic knowledge in diffusion models.

🛠️TOP TOOLS

Reading Coach - a tool designed to help improve reading skills. It offers personalized assistance and feedback, focusing on enhancing reading fluency and comprehension.

ArtificialAnalysis.ai - provides comparative analysis of various AI models and hosting providers, focusing on model quality, speed, and pricing. provides comparative analysis of various AI models and hosting providers, focusing on model quality, speed, and pricing. 

Broadcast - AI-driven tool for managers, enhancing team productivity by automatically taking meeting notes, tracking decisions, and creating action items.

Interflexion - AI-powered role-play coaching, focusing on developing leadership and interpersonal skills. It provides personalized coaching, microlearning scenarios, and professional curriculum development, supported by AI for insightful feedback. 

VoiceSpin - AI-powered, cloud-based contact center solutions for both sales and customer support teams. It features tools like AI auto dialer, speech analyzer, and bulk SMS, along with VoIP telephony. 

🗞️MORE NEWS

How Beloved Indie Blog 'The Hairpin' Turned Into an AI Clickbait Farm

The beloved indie blog 'The Hairpin,' known for its unique and intimate content, has unfortunately turned into an AI clickbait farm. Originally shut down in 2018, it was revived this year, but now it's filled with low-quality AI-generated articles designed to drive search engine traffic. The site's transformation is emblematic of the challenges facing digital media, with original articles altered and author bylines replaced. The Hairpin's new owner, a Serbian DJ, admits the majority of posts are AI-generated, focusing on SEO rather than quality content. This situation serves as a cautionary tale for other media properties and highlights the evolving landscape of online journalism and content creation. WIRED

Ola founder’s Krutrim becomes India’s first AI unicorn

Krutrim, an AI startup founded by Ola's Bhavish Aggarwal, has become India's first AI unicorn, reaching a valuation of $1 billion. This achievement makes it the fastest unicorn in India, achieved just a year after its founding. Matrix Partners India led the $50 million funding round. Krutrim is developing a large language model trained on Indian languages and English, aiming to launch a multilingual conversational AI assistant. Additionally, the startup plans to release APIs for developers and enterprises, and is exploring chip manufacturing optimized for AI computing. Aggarwal, also behind Ola Electric, emphasized the importance of India developing its own AI capabilities. TECHCRUNCH

Chef Robotics eyes commercial kitchens with $14.75M raise

Chef Robotics, a company specializing in kitchen automation, has raised $14.75 million in a combined equity and debt funding round, bringing its total funding to $22.5 million. This San Francisco-based startup, established in 2019, focuses on food assembly automation, differentiating itself from competitors like Miso, which is known for its burger-cooking robotic arm, Flippy. Chef Robotics is implementing a Robotics as a Service (RaaS) model to make their technology more accessible to businesses without the means to afford high upfront costs. The funding will be used to enhance the company's go-to-market strategy, hire more engineers and technicians, and further develop ChefOS, the software that powers its robots. TECHCRUNCH

Europe has an AI champion—the Dutch semiconductor giant that is the continent’s most valuable tech firm and just hit a record high

ASML Holding NV, Europe's most valuable tech firm and a key player in the AI boom, saw its stock reach a record high this week, indicating strong demand for its semiconductor equipment. The company, known for its advanced extreme ultraviolet (EUV) lithography systems essential in modern chip manufacturing, experienced a significant rebound in orders, particularly from memory chipmakers. This surge in demand is attributed to the growing need for high-performance memory chips, crucial for realizing the full potential of AI. ASML's CEO, Peter Wennink, acknowledges this need as a driving force behind the increased bookings for their EUV systems, which have a delivery time of 12 to 18 months. FORTUNE

Satya Nadella says the explicit Taylor Swift AI fakes are ‘alarming and terrible’

Microsoft CEO Satya Nadella has expressed concern over the creation and spread of sexually explicit AI-generated images of Taylor Swift, calling the situation "alarming and terrible." In an upcoming NBC Nightly News interview, Nadella acknowledges the need for swift action and emphasizes the importance of implementing more robust guardrails around AI technology to ensure safer content production. THE VERGE