Today:

Elon Musk's Statement SHOCKs the Entire Reality! AGI, living in a simulation, Groq, Q-Star and Gemma

Elon Musk sparked a fascinating chat on Twitter about AGI (Artificial General Intelligence) and its potential to prove or disprove if we're living in a simulation. He tossed around the idea that if AGI could unearth new physics laws, it might hint we're in a "base layer" of reality.

Meanwhile, the tech world is buzzing with AI developments, from Nvidia's AI research hinting at robots that could bridge the gap between simulations and real-world applications, to the intrigue of simulation technology being a data goldmine. Simulations, seen as data wells, are invaluable for advancing AI, robotics, and even understanding the fabric of our reality. The concept of creating detailed simulations for training AI, alongside Musk's simulation theory, underscores the blurry line between digital constructs and physical existence. 

WATCH THE VIDEO ON YOUTUBE

Chrome’s new AI feature can help you write on the web

Chrome's latest update, M122, introduces "Help me write," an experimental AI feature aimed at assisting users with online content creation. Available on Mac and Windows PCs in the U.S., it utilizes Gemini models to suggest and refine text based on the context of the webpage being viewed. 

Whether crafting a review, selling items, or making inquiries, the tool provides relevant suggestions to enhance writing. Users can activate this feature in Chrome settings and access it by right-clicking on a text field. With its user-friendly interface, "Help me write" simplifies everyday writing tasks, such as booking hotel reservations.

GOOGLE

Google upgrades AI product for advertisers with Gemini models

Google announced the integration of its Gemini AI models into its ad product, enhancing ad placements across its services. This move aims to expand AI capabilities to more users, particularly in generating conversational text and realistic images. Gemini will upgrade Performance Max, allowing longer ad headlines and image generation featuring people. 

Google ensures safety measures to prevent misuse, such as watermarking synthetic images. As generative AI usage rises, concerns about misinformation grow, prompting Google to implement safeguards. This advancement reflects Google's commitment to AI innovation while addressing ethical considerations in content creation and dissemination.

REUTERS

Google cut a deal with Reddit for AI training data

Google and Reddit have struck a deal worth $60 million annually, granting Google access to Reddit's real-time data for AI training. The partnership enables Google to enhance its search capabilities using Reddit's vast content. Reddit, in turn, gains access to Google's Vertex AI service. 

Despite past tensions, this collaboration aims to mutually benefit both parties. Reddit's move aligns with its IPO plans, potentially boosting its valuation. This agreement reflects the growing importance of data partnerships in AI development, highlighting the value of real-time, user-generated content for training AI models.

THE VERGE

Google to fix AI picture bot after 'woke' criticism

Google is scrambling to fix its new AI image tool after facing criticism for overemphasizing diversity at the expense of historical accuracy. Users reported that the Gemini bot supplied incorrect images, such as depicting women and people of color when asked for images of America's founding fathers. 

The company acknowledged the issue, stating the tool was "missing the mark" and pledged to improve. Google suspended the tool's ability to generate images of people while working on a fix. This incident echoes past AI blunders around diversity, including Google's infamous labeling of a black couple as "gorillas" in its photos app.

BBC

Reddit says it’s made $203M so far licensing its data

Reddit, gearing up for its stock market debut, highlights its lucrative data licensing deals with AI firms like OpenAI in its IPO filing. These agreements, worth $203 million as of January 2024, involve training AI models on Reddit's vast content repository. CEO Sam Altman's ties to both Reddit and OpenAI suggest a close relationship between the two. 

Reddit's decision to monetize its data stems from the recognition of its value in AI model training. As AI adoption rises, content producers seek similar licensing deals to offset potential traffic losses to AI-powered platforms and mitigate legal risks.

TECHCRUNCH

Microsoft and Intel strike a custom chip deal that could be worth billions

Microsoft and Intel struck a deal worth over $15 billion, where Intel will make custom chips for Microsoft's use. The chips, designed by Microsoft, will utilize Intel's 18A process. This partnership marks Intel's push to reclaim its chipmaking dominance, a strategy similar to TSMC's success with partnerships. 

The move comes amid rising demand for self-designed chips, but Intel faces challenges, including delays in opening a new chip plant in Ohio. Microsoft CEO Satya Nadella expressed excitement about the transformative potential of the collaboration.

THE VERGE

Stable Diffusion 3 arrives to solidify early lead in AI imagery against Sora and Gemini

Stability AI rolls out Stable Diffusion 3, beefing up its image-generating AI against rivals like Sora and Gemini. With upgraded architecture and flexibility across hardware, SD3 promises scalability and quality. The model suite varies from 800 million to 8 billion parameters, catering to diverse needs. Notably, it emphasizes multimodal understanding and video capabilities, setting it apart. Safety measures are highlighted, aiming to prevent misuse. 

While specifics are sparse, Stable Diffusion aims to dominate the generative AI landscape, prioritizing utility over exclusivity. Expectations are high, awaiting further technical insights and real-world performance.

TECHCRUNCH

🧠RESEARCH

LongRoPE: Extending LLM Context Window Beyond 2 Million Tokens

 LongRoPE is a groundbreaking method that expands language models' context window to over 2 million tokens, significantly beyond the current 128k limit, with minimal fine-tuning. This leap is made possible through innovative techniques addressing positional embedding challenges, allowing for broader context understanding in tasks without losing performance on shorter texts. This advancement promises to significantly enhance the capabilities of language models in processing and understanding long documents. 

Aria Everyday Activities Dataset

Aria Everyday Activities Dataset featuring 143 daily activity sequences captured through Project Aria glasses. Recorded in diverse indoor locations by various wearers, it includes multimodal sensor data like 3D trajectories, scene point clouds, eye gaze vectors, and speech transcriptions. This rich dataset supports research in neural scene reconstruction and segmentation, offering open access and tools for enhanced machine perception studies. 

SDXL-Lightning: Progressive Adversarial Diffusion Distillation

SDXL-Lightning is a new method in text-to-image generation that sets the bar higher with state-of-the-art results. By blending progressive and adversarial distillation techniques, it achieves a sweet spot between image quality and variety. The discussion covers theoretical underpinnings, discriminator design, model formulation, and training strategies. The authors also share the distilled SDXL-Lightning models for wider use.

User-LLM: Efficient LLM Contextualization with User Embeddings

User-LLM, a new way to make big language models (like the ones that understand and generate text) smarter by using info about how users interact with them. This method creates a special profile for each user from their past actions to help the model get what the user might want next. It's been tested and works better than older methods, especially for tricky tasks or when dealing with lots of info, and it doesn't need as much computer power.

YOLOv9: Learning What You Want to Learn Using Programmable Gradient Information

YOLOv9 is a cutting-edge approach in deep learning that tackles the challenge of information loss in deep networks. It presents a novel concept called Programmable Gradient Information (PGI) to preserve complete input data through the network, ensuring better learning outcomes. Alongside, it showcases a new lightweight network architecture, Generalized Efficient Layer Aggregation Network (GELAN), demonstrating superior results on the MS COCO dataset for object detection without relying on complex operations. This technique promises improved model performance with efficient parameter use, outperforming traditional models even when starting from scratch. 

🛠️TOP TOOLS

Artbox - offers a library of high-quality AI-generated visuals for projects, delivering amazing visuals at no cost. 

AgentHub - offers a platform for integrating and managing various AI agents from different providers, aiming to streamline the development of AI-powered applications.

Shortform-  offers in-depth guides and summaries of non-fiction books, covering a vast array of topics.

Deciphr - Automate your podcast content workflow

Outset - AI-moderated video interviews, offering the rapidity of surveys with the depth of personal interviews, all captured in video format. 

📲SOCIAL MEDIA

🗞️MORE NEWS

Nvidia Declares AI a ‘Whole New Industry’—and Investors Agree

Big tech's AI investments soar, with Nvidia leading the charge. CEO Jensen Huang declares AI as a booming sector, attributing Nvidia's growth to generative AI. Investors back the claim, propelling Nvidia's stock after impressive earnings, signaling continued AI spending. AI emerges as a lucrative industry, driving Nvidia's success. THE WALL STREET JOURNAL

Jasper acquires Stability AI’s Clipdrop to strengthen marketing copilot

Jasper, a SF startup known for AI-driven marketing, acquires Clipdrop for image editing. The deal aims to enhance Jasper's marketing copilot, enabling personalized ad strategies. Clipdrop's platform allows image creation and editing, complementing Jasper's AI capabilities. The move reflects the growing demand for AI-powered creative tools in marketing. VENTUREBEAT

Nvidia has begun sampling two new AI chips for China market, says CEO

Nvidia CEO announces sampling of two new AI chips for China amid US export restrictions. The move aims to maintain market dominance despite declining sales due to export controls. Nvidia's quarterly sales in China dropped significantly, prompting the company to seek customer feedback on the new chips designed to comply with regulations. REUTERS

Samsung is bringing Galaxy AI features to more devices

Samsung announces One UI 6.1 update bringing Galaxy AI features to more devices including Galaxy S23 series, Z Fold5, and Tab S9 Ultra. Features include Google's "Circle to Search," Live Translate, Chat Assist, Note Assist, Browsing Assist, Generative Edit, and more, enhancing user experience and accessibility. TECHCRUNCH

Newo.ai’s ‘digital employee’ merges AI agent with robot

Newo.ai merges AI agents with physical robots, creating "digital employees" for office tasks. Co-founded by David Yang, it offers a platform to replicate human intelligence for various roles. With customizable programming, it suits tasks from reception to technical support. Customers include spas and fitness centers, finding it cost-effective and efficient. VENTUREBEAT

AI billionaire crowns ‘whole new industry’ amid ‘tipping point’ after his company passes Google for third most valuable in the world

Jensen Huang, CEO of Nvidia, celebrates as his company surpasses Google in valuation. Nvidia's surge is attributed to its pivotal role in the AI revolution, with Huang emphasizing the shift to "accelerated computing." Nvidia's AI tokens are deemed essential for various tech applications, propelling its growth amidst high expectations. FORTUNE

The Justice Department gets a chief AI officer

The Justice Department appoints Princeton professor Jonathan Mayer as its first chief AI officer to navigate AI's impact on law enforcement. Attorney General Merrick Garland emphasizes the need to keep pace with technological advancements. Mayer will lead a team focusing on cybersecurity and AI, aligning with Biden administration's directives on AI talent and fairness standards. THE VERGE