Today:

• Altman Funds Clean Energy Tech

• Apple Secures French AI Startup

• SoftBank Boosts AI with $960M

• Qualcomm Partners with G42's Core42

• Meta's Taylor Moves to Microsoft AI

• Hollywood's CAA Tests AI Digital Replicas

Sam Altman Invests in Energy Startup Focused on AI Data Centers

The AI industry's booming data centers are hungry for power, facing a major challenge in meeting their electricity needs. To address this, tech mogul Sam Altman and venture capital firm Andreessen Horowitz are investing $20 million in Exowatt. This startup aims to revolutionize clean energy for large data centers, proposing a novel approach combining solar power and energy storage. 

With global data center capacity projected to double by 2030, solutions like Exowatt's are crucial for sustaining the AI revolution without overtaxing the power grid. This investment underscores the growing importance of energy efficiency in the tech sector's expansion.

THE WALL STREET JOURNAL

Apple acquires French startup behind AI and computer vision technology

Apple has reportedly acquired Datakalab, a French startup specializing in AI compression and computer vision tech. The deal, finalized in December, strengthens Apple's on-device AI capabilities. Datakalab's expertise in low-power, efficient deep learning aligns with Apple's focus on privacy and speed. The startup's advanced vision technology may enhance Apple's future products like iOS 18 and Vision Pro, potentially impacting features like facial recognition in Photos and Face ID. 

While neither company has confirmed, reports suggest Datakalab's founders didn't join Apple, but several employees did. The move reflects France's growing influence in AI development.

9TO5MAC

SoftBank to spend $960m to boost computing power for generative AI

SoftBank, the Japanese telecom giant, plans to invest $960 million by 2025 to bolster its computing capabilities for developing cutting-edge generative AI. This move aims to position SoftBank among Japan's leaders in computing power and to pioneer a Japanese-language-specific generative AI with top-notch performance. 

The investment underscores SoftBank's commitment to advancing AI technology and reflects Japan's push to stay competitive in the global AI landscape.

NIKKEI

UAE-based AI firm G42 announces collaboration with U.S. group Qualcomm

UAE-based AI firm G42 has announced a collaboration with Qualcomm, the US technology giant. G42's subsidiary, Core42, will integrate Qualcomm's Cloud AI 100 products into its Condor AI platform. 

This partnership aims to enhance G42's AI capabilities, particularly in the realm of generative artificial intelligence. The move highlights the growing synergy between international AI players and signifies Qualcomm's importance in powering advanced AI solutions.

REUTERS

Microsoft hires former Meta exec to bolster AI supercomputing team

Former Meta VP of Infrastructure, Jason Taylor, joins Microsoft’s AI team as Corporate VP and Deputy CTO. Microsoft aims to advance AI systems with robust hardware. Taylor’s experience includes managing AI, data, and privacy infrastructure at Meta. He also chaired the Open Compute Project Foundation, promoting open-source designs for data centers. 

Rumors of a $100 billion supercomputer, "Stargate," were refuted by Microsoft's CTO. This move follows Microsoft’s recent hiring of Google DeepMind co-founder Mustafa Suleyman to lead a new consumer AI division overseeing Copilot, Bing, and Edge.

THE VERGE

Hollywood Talent Agency CAA Tests AI Clones

Creative Artists Agency (CAA), known for representing Hollywood stars, is exploring AI's role in entertainment. With concerns over AI replacing jobs, CAA introduced CAA Vault, allowing celebrities to create digital replicas. This move reflects Hollywood's mixed feelings about AI's impact on creativity and employment. 

The initiative aims to find ways for entertainers to benefit from technological advancements rather than fear them. CAA's approach exemplifies Hollywood's ongoing adaptation to evolving technologies, seeking to harness AI while navigating its potential consequences for the industry's future.

THE INFORMATION

🧠RESEARCH

TextSquare: Scaling up Text-Centric Visual Instruction Tuning

The paper discusses advancing text-centric visual question answering (VQA) with Multimodal Large Language Models (MLLMs). They introduce Square-10M dataset, generated using closed-source MLLMs, improving model TextSquare's performance. TextSquare surpasses open-source models and rivals top-tier ones like GPT4V and Gemini, demonstrating the importance of extensive instruction-tuning data.

Groma: Localized Visual Tokenization for Grounding Multimodal Large Language Models

Groma is a Multimodal Large Language Model (MLLM) with advanced visual understanding, focusing on region-level tasks like region captioning and visual grounding. Groma employs localized visual tokenization, breaking images into regions for precise analysis. It outperforms models without localization in referring and grounding tasks, showcasing the benefits of integrated localization.

PhysDreamer: Physics-Based Interaction with 3D Objects via Video Generation

PhysDreamer is a method for creating realistic interactions with 3D objects in virtual environments. Unlike traditional approaches, PhysDreamer incorporates physics-based dynamics by leveraging learned priors from video generation models. This enables objects to respond realistically to novel interactions, enhancing immersion in virtual experiences. PhysDreamer addresses the challenge of estimating physical material properties, paving the way for more engaging virtual environments.

Does Gaussian Splatting need SFM Initialization?

The paper examines the necessity of Structure-from-Motion (SFM) initialization for 3D Gaussian Splatting, a method used for scene reconstruction and view synthesis. It explores alternative initialization strategies and proposes utilizing Neural Radiance Fields (NeRF) for volumetric reconstructions to overcome SFM dependency. Results show improved or equivalent performance compared to SFM initialization.

AutoCrawler: A Progressive Understanding Web Agent for Web Crawler Generation

AutoCrawler is a framework that combines large language models (LLMs) with crawlers to enhance web automation. Traditional methods like wrappers face limitations in adaptability and scalability. AutoCrawler utilizes a two-stage approach, leveraging HTML structure for progressive understanding and efficient action generation. Experiments demonstrate its effectiveness in handling diverse web environments.

🛠️TOP TOOLS

Instashorts - TikTok stories in your voice!

Metaforms - AI Forms for Feedback, Surveys and Research

Apollo - Find and close your next deal fueled by the most accurate data on 275M contacts and 73M companies

Llm.c - LLM training in simple, raw C/CUDA

Plandex AI - An open source, terminal-based AI coding engine that helps you complete large tasks, work around bad output, and maximize productivity

📲SOCIAL MEDIA

🗞️MORE NEWS

Alethea AI launches emotive and expressive AI avatars on Coinbase blockchain

Alethea AI launches lifelike AI avatars on Coinbase blockchain, revolutionizing digital interactions. With its Emote Engine, users can craft expressive characters embodying a range of emotions. This user-friendly platform empowers creators, fostering a new era of AI-driven creativity and economic opportunities. Integration with blockchain enables direct monetization and transparent transactions, marking a significant leap in avatar technology. VENTUREBEAT

Mega Networks plans to make AI servers in Maharashtra

Mega Networks, chosen for India's PLI scheme, plans to manufacture AI servers in Maharashtra. With an initial investment of Rs 350-400 crore, the factory aims to address the growing demand for AI servers. The company expects significant revenue growth and intends to expand its workforce to meet market demands. ECONOMICS TIMES

AMD teams up with Arm to unveil AI chip family that does preprocessing, inference and postprocessing on one silicon — but you will have to wait more than 12 months to get actual products

AMD collaborates with Arm to unveil Versal AI Edge Series Gen 2 chips, integrating preprocessing, AI inference, and postprocessing in one silicon. Promising end-to-end acceleration for AI-driven embedded systems, the chips aim at diverse markets like autonomous driving and healthcare, boasting improved safety and performance. Availability is expected in late 2025. TECHRADAR

BBC to invest £6m in AI to help transform its education services

The BBC plans to invest £6 million in AI to enhance its educational platform, BBC Bitesize, aiming to personalize learning and engage young audiences. The move comes after the platform's crucial role during the pandemic. New features will include personalized testing and tailored content recommendations, aligning with the BBC's mission to inform, educate, and entertain. THE GUARDIAN

An AI Star Seeks to Bring Self-Driving Cars to Japan by 2030

Issei Yamamoto, famed for his AI's victory in Japanese chess, now aims to pioneer self-driving cars in Japan by 2030. Backed by major firms like Mizuho Financial Group and NTT Docomo Ventures, his startup Turing Inc. secured $19.4 million in funding to develop AI models and vehicles for this ambitious project. BLOOMBERG

Google builds an AI model that can predict future weather catastrophes

Google introduces SEEDS, an AI model for weather forecasting, claiming to generate accurate predictions faster and cheaper than traditional methods. SEEDS utilizes generative AI to produce multiple weather scenarios efficiently, addressing challenges in predicting extreme weather events and enhancing forecasting accuracy. LIVE SCIENCE

AI Detects Mysterious Detail Hidden in Famous Raphael Masterpiece

An AI neural network has detected a peculiar detail in a Raphael masterpiece, the Madonna della Rosa: St Joseph's face likely wasn't painted by Raphael. Developed by UK and US researchers, the AI algorithm trained on Raphael's style suggests the face differs from the rest, shedding new light on the painting's authenticity. SCIENCE ALERT

Bootstrapped AI startup Smallstep serves up Misal, a Marathi LLM

Smallstep.ai, a Bengaluru-based startup, introduces Misal, an AI model for Marathi. Named after a popular Maharashtrian dish, Misal aims to enhance AI capabilities in local languages. Developed on Meta’s Llama2 model, Misal outperforms ChatGPT 3.5 in reading but lags in sentiment analysis and translation. MONEY CONTROL

Ex-Amazon exec claims she was asked to ignore copyright law in race to AI

A former Amazon AI scientist alleges discrimination and copyright breaches in a lawsuit. Viviane Ghaderi claims demotion and dismissal post-pregnancy, citing pressure to ignore copyright rules for AI advancement. Amazon denies wrongdoing, but faces a trial over sex discrimination, retaliation, and wrongful termination charges. Ghaderi seeks justice for her treatment. THE REGISTER

How United Airlines uses AI to make flying the friendly skies a bit easier

United Airlines leverages AI across its operations to enhance customer experience and streamline internal processes. From generating flight delay messages to coordinating crew assignments, AI plays a significant role. CIO Jason Birnbaum emphasizes the importance of AI in optimizing cloud usage and improving operational efficiency. TECHCRUNCH