Today:

Google's Genie SHOCKS the Industry | AI Creates Unlimited Playable Games | Foundation World Model.

Google unveils Genie, a foundation world model, learns from vast internet video data to create interactive 2D worlds from image prompts. This marks a trend of accelerating breakthroughs in AI. 

With Genie's ability to simulate varied realities, the path to AGI becomes clearer. Its unsupervised learning and action-controllable capabilities pave the way for training versatile AI agents. 

WATCH THE VIDEO ON YOUTUBE

Apple cancels work on electric car

Apple has scrapped its electric car project after a decade of effort, redirecting employees to AI development. Market trends, including high interest rates and subdued demand for electric vehicles, led to this decision. The company's initial plan for a self-driving car evolved into a more conventional vehicle with advanced driver-assistance features. 

Despite previous reports suggesting a 2024 or 2025 release, progress was uneven, with layoffs in 2019 and design changes. The move reflects broader shifts in the automotive industry, with other major players also scaling back EV investments amid changing consumer sentiment.

REUTERS

Tumblr and Wordpress to Sell Users’ Data to Train AI Tools

Tumblr and Wordpress are set to sell user data to Midjourney and OpenAI, as revealed by internal documents. The specifics of the data and recipients are unclear, but it's evident from internal communications that deals between Automattic, the parent company, and OpenAI and Midjourney are imminent. 

Internal turmoil within Tumblr is highlighted, with a product manager mentioning an inadvertent compilation of user posts for OpenAI and Midjourney. Whether this data has been sent or not remains ambiguous. The situation underscores concerns over privacy and ethical data usage.

404 MEDIA

Apple and Tesla Veterans Aim to Help Japan Design AI Chips

Japan's semiconductor research group teams up with US startup Tenstorrent Inc. to craft an advanced AI chip, marking Japan's dive into the AI chip market. Led by industry veterans from Tesla and Apple, Tenstorrent will contribute its design expertise to Japan's AI accelerator chip.

 The collaboration aims to offer an alternative to dominant players like Nvidia and Arm Holdings, using the open-source RISC-V standard. This move signals Japan's intent to carve its niche in AI hardware development, leveraging expertise from both nations.

BLOOMBERG

Microsoft’s GitHub Offers Companies Souped-Up AI Coding Tool

Microsoft's GitHub launches GitHub Copilot Enterprise, a pricier version of its AI coding tool. Costing $39 per user per month, it aims to aid engineers in understanding and working with their own code. The tool utilizes AI to answer queries based on the company's programming code, facilitating faster coding and onboarding for new team members. 

GitHub seeks to attract more subscribers by integrating AI features, enhancing collaboration, issue resolution, and program modernization. This move underscores GitHub's commitment to leveraging AI to streamline software development processes and empower coders.

BLOOMBERG

GoDaddy launches a suite of AI tools for small businesses

GoDaddy launched Airo, a new set of AI-driven marketing tools for small businesses. Airo offers domain name suggestions, logo generation, website setup, social media assistance, email marketing, and ad placement. The AI optimizes content for search engines and aims to streamline online presence management. Similar all-in-one platforms include Mailchimp, Squarespace, and Wix, all leveraging AI for tasks like email copywriting, content generation, and website building. 

These services reduce the need for multiple software vendors and outside consultants, easing the burden on small business owners and enhancing their online presence without the need for extensive creative direction.

FAST COMPANY

🧠RESEARCH

ChatMusician: Understanding and Generating Music Intrinsically with LLM

"ChatMusician" integrates musical abilities into Large Language Models (LLMs), bridging text and music via ABC notation. It surpasses GPT-4 in composing structured music, even outperforming on a music understanding benchmark. The model, devoid of external structures, showcases potential for music generation, as evidenced by MusicPile and MusicTheoryBench datasets released on GitHub.

FuseChat: Knowledge Fusion of Chat Models

FuseChat enhances large language models (LLMs) by fusing knowledge from diverse existing LLMs, reducing the need for expensive individual training. Unlike direct parameter blending, FuseChat utilizes lightweight continual training to merge multiple chat LLMs, resulting in a more robust model. Experimental validation shows FuseChat's superiority across various chat domains, even outperforming GPT-3.5.

Multi-LoRA Composition for Image Generation

The paper introduces methods for effectively combining multiple Low-Rank Adaptations (LoRAs) in text-to-image models to enhance image synthesis. Two training-free approaches, LoRA Switch and LoRA Composite, are proposed, showing clear performance improvements over baseline methods, especially with an increased number of LoRAs. Evaluation is conducted using the ComposLoRA testbed.

StructLM: Towards Building Generalist Models for Structured Knowledge Grounding

StructLM aims to enhance large language models' (LLMs) proficiency in interpreting structured data like tables and graphs. They introduce a comprehensive instruction tuning dataset and train models based on the Code-LLaMA architecture, ranging from 7B to 34B parameters. StructLM outperforms task-specific models on various structured knowledge grounding tasks, showcasing exceptional generalization but suggesting that scaling model size offers limited benefits.

Nemotron-4 15B Technical Report

The Nemotron-4 15B technical report unveils a 15-billion-parameter multilingual language model trained on 8 trillion text tokens. It excels in English, multilingual, and coding tasks, outperforming other similarly-sized models in four out of seven evaluation areas and showing competitive performance in the rest. Notably, it surpasses larger models and those explicitly designed for multilingual tasks in multilingual capabilities.

MegaScale: Scaling Large Language Model Training to More Than 10,000 GPUs

"MegaScale" details the architecture and deployment of a production system capable of training large language models (LLMs) on over 10,000 GPUs. Addressing challenges in efficiency and stability, the system employs a full-stack approach, optimizing algorithms, system components, and data pipelines. Achieving a 55.2% Model FLOPs Utilization (MFU), it improves efficiency by 1.34x compared to previous systems like Megatron-LM, with a focus on fault tolerance and straggler mitigation. The paper aims to inspire future research in LLM systems.

🛠️TOP TOOLS

Venture GPT - Co-pilot for VC and startups.

Findr -  AI-powered search tool that streamlines workplace productivity by enabling users to search across multiple apps with a single query. 

izTalk - AI-powered solutions for customer engagement, offering tools like chatbots and virtual assistants to improve communication and streamline business operations.

Guidde - uses AI to create video documentation, simplifying the process of making how-to guides, onboarding materials, and FAQs, making it faster and easier for businesses.

Never - AI-powered photo app that creates realistic images and avatars from your photos. It's user-friendly and offers a variety of styles for social media enhancement. 

📲SOCIAL MEDIA

🗞️MORE NEWS

OpenAI claims the Times cheated to get ChatGPT to regurgitate articles

OpenAI accuses The New York Times of using sneaky tactics to make ChatGPT regurgitate its articles, seeking to axe some copyright claims. The AI firm argues the Times exploited a bug and fed the bot articles directly, unlike typical users. The lawsuit underscores the growing legal fracas around AI technology. THE VERGE

Google Is Paying Publishers to Test an Unreleased Gen AI Platform

Google pays news publishers for early access to a new AI platform, aiming to gather feedback. Publishers must generate content using the tool, which summarizes articles from various sources. Concerns arise over potential plagiarism and negative impact on original sources. The program is part of Google's News Initiative, facing mixed reception. ADWEEK

Photoroom, An App That Generates AI Images In One Second, Is Now Worth $500 Million

Photoroom, a Paris-based startup, has achieved a $500 million valuation after securing $43 million in Series B funding. Launched in 2019, the AI-based photo editing app boasts 30 million users worldwide, enabling them to enhance product photos with AI-generated backgrounds and text-to-image AI. The startup aims to democratize high-quality photography for businesses and individuals, with notable clients including Netflix and Warner Bros. Their latest foundational model, Photoroom Instant Diffusion, promises lightning-fast image generation in just one second, setting them apart in the market. FORBES

Abu Dhabi’s Largest Firm Adds AI-Powered Observer to Board

Abu Dhabi's International Holding Co. is boosting its board with an AI-powered observer named Aiden Insight. Valued at $238 billion, the move aims to enhance strategic decisions by processing decades of business data, financials, and market trends. This step underscores Abu Dhabi's deepening engagement with artificial intelligence. BLOOMBERG

Samsung unveils new memory chip with ‘highest-capacity to date’ for AI

Samsung Electronics has unveiled a cutting-edge high-bandwidth memory chip, HBM3E 12H, boasting the industry's highest capacity yet. The chip promises over 50% performance and capacity enhancement, catering to the burgeoning demand from AI service providers. Samsung aims for mass production by mid-2024, anticipating a positive market response. CNBC

Intel aims to deliver chips for 100 million AI PCs by 2025

Intel is targeting to provide processors for 100 million AI-enabled PCs by 2025, aiming to tap into the next wave of growth in the tech industry. With plans to deliver 40 million AI PCs this year and 60 million next year, Intel anticipates capturing over 20% of the global PC market by 2025. NIKKEI

AI Startup Exodigo Raises $105 Million to Make Underground Maps

Exodigo, an AI startup specializing in underground mapping, secures $105 million in Series A funding led by Greenfield Partners and Zeev Ventures. New investors include SquarePeg, 10D VC, and National Grid Partners. The Tel Aviv and Palo Alto-based company plans to use the capital to enhance its products, expand its team, and grow globally. BLOOMBERG

Microsoft now offers Copilot GPTs to help you work out, find recipes, and more

Microsoft introduces new Copilot GPTs for fitness, design, travel planning, and cooking. Users can access tailored assistance and will soon have the ability to create their own Copilot GPTs, as announced by Microsoft's corporate vice president, Jordi Ribas. THE VERGE

Alibaba Leads Record Deal to Mint $2.5 Billion China AI Firm

Alibaba spearheads a record-breaking $2.5 billion funding round for Chinese AI startup Moonshot AI, alongside existing investor Monolith Management. The investment underscores Alibaba's strategic focus on generative AI, aligning with trends in the tech industry. Other backers include Tencent and Silicon Valley giants like Microsoft, reflecting growing interest in AI technologies. BLOOMBERG

AI startup Glean lures Citigroup as investor at $2.2 billion valuation after revenue quadruples

Glean, an AI startup valued at $2.2 billion, attracts investment from Citigroup and tech firms like Databricks and Workday. Annualized revenue reaches $39 million, up from $10 million a year earlier. Glean's AI assistant competes with Microsoft's Copilot, offering data-driven answers to workplace queries. Expansion plans include diversifying into financial services, retail, and manufacturing sectors. CNBC

Qualcomm’s Newest Chip Brings AI to Wi-Fi

Qualcomm has introduced the FastConnect 7900 chip, leveraging AI to enhance Wi-Fi 7 connectivity. This chip aims to optimize power and latency on a per-application basis by detecting the specific applications being used, thus improving reliability in wireless connections. The FastConnect 7900 also integrates Wi-Fi, Bluetooth, and ultra-wideband technologies, simplifying complexity for consumers. Additionally, the chip supports hosting two Wi-Fi connections on the same spectrum band, further enhancing connectivity options. Qualcomm is already sampling the chip to manufacturers, with products expected to hit the market in the second half of the year. IEEE SPECTRUM

Microsoft's deal with Mistral AI faces EU scrutiny

Microsoft's investment in French AI startup Mistral AI sparks EU scrutiny, with concerns over tech giant's growing power. Lawmakers call for investigation into partnership, amid ongoing antitrust scrutiny over Microsoft's ties with OpenAI. Questions raised over Mistral's lobbying efforts for AI Act exemptions, seen as protecting European startups. EU urged to act swiftly. REUTERS

AI designs bespoke 3D-printed prosthetic eyes

Using AI and 3D printing, prosthetic eyes can now be made faster, reducing production time by 80%. This innovation, developed by researchers in Germany, promises more comfortable fittings, especially for children. With potential to ease access, it marks a significant advancement in ocular prosthetics. NEWSCIENTIST

Putting AI into the hands of people with problems to solve

Pienso, co-founded by MIT alumni, offers a no-code AI builder for non-experts, empowering them to develop machine learning models without coding. Originating from a project on cyberbullying, Pienso's platform aids in various domains like misinformation detection and disease analysis, fostering AI solutions tailored by those who understand the data best. MIT