Today:

Elon Musk to Open Source Grok AI | "Murdering AI, Cloning Humans & AI Gods" OpenAI's Scott Aaronson

Elon Musk announces that XAI, the company behind Grok, will open-source Grok this week. No specific details yet on what aspects will be open-sourced. Musk advocates for more companies to move in this direction, amidst his recent lawsuit with OpenAI. 

Scott Aaronson, a theoretical computer scientist, reflects on AI safety and its ethical implications. He ponders on the uniqueness of human creativity and the potential consequences of AI surpassing human capabilities. The debate continues on whether AI should be allowed to reach the pinnacle of creative genius and the ethical implications of AI's ability to replicate human consciousness. 

WATCH THE VIDEO ON YOUTUBE

Midjourney bans all Stability AI employees over alleged data scraping

Midjourney, a generative AI company, has banned Stability AI employees from its service, alleging they caused a 24-hour outage by scraping data. Midjourney linked the outage to "botnet-like activity" from paid Stability AI accounts. Stability AI CEO Emad Mostaque denies ordering the actions and is investigating. 

Midjourney founder David Holz provided information to aid the investigation. The incident highlights concerns over data scraping and copyright issues in the AI community. Both companies have faced criticism and lawsuits for using online data without consent. As the situation develops, no further updates have been provided.

THE VERGE

Jeff Bezos’s investment in Perplexity AI has nearly doubled in value in a few months as Google challenger nears $1B unicorn status

Jeff Bezos's investment in Perplexity AI, a Google competitor, has nearly doubled in value within months, edging the startup closer to a $1 billion unicorn status. Despite historical failures in challenging Google, Bezos's backing indicates confidence in Perplexity AI's potential. 

The investment underscores the growing interest in AI technologies and their impact on the competitive landscape.

FORTUNE

Abu Dhabi Targets $100 Billion AUM for AI Investment Firm

Abu Dhabi is launching MGX, a tech investment firm focused on AI and semiconductors. Backed by Mubadala Investment Co. and G42, the firm aims to manage over $100 billion in assets. 

Ahmed Yahia Al Idrissi, CEO of Mubadala's direct investments platform, will lead MGX. This move underscores Abu Dhabi's commitment to tech innovation and strategic investments in cutting-edge sectors.

BLOOMBERG

Nvidia Says NeMo AI Platform Complies With Copyright After Authors’ Complaint

Nvidia reassures that its NeMo AI platform complies with copyright law after authors filed a complaint. The authors allege Nvidia used their books without permission to train its AI models. 

The complaint, filed in California, pits Nvidia against Abdi Nazemian, Brian Keene, and Stewart O’Nan, who claim their copyrighted works were used in NeMo.

THE WALL STREET JOURNAL

Pika adds generative AI sound effects to its video maker

Pika, a player in the AI video space, introduces a new feature allowing users to generate sound effects for their AI videos directly within the platform. This addition complements its existing lip-syncing capabilities, making it an all-in-one AI video creation platform. 

Users can generate sound effects through contextual generation or by providing specific prompts. Initially available to select users, Pika plans to roll out this feature gradually to its platform users. This move positions Pika as a strong competitor against other players in the creative AI space.

VENTUREBEAT

🧠RESEARCH

ELLA: Equip Diffusion Models with LLM for Enhanced Semantic Alignment

ELLA is a tool that upgrades text-to-image generation by integrating large language models (LLMs) for better understanding complex prompts. ELLA connects pre-trained models using a novel Timestep-Aware Semantic Connector, enhancing the creation of images from detailed descriptions without retraining. It shines in handling prompts with multiple objects and intricate details, proven by tests on a new, tough benchmark, DPG-Bench. The results show ELLA's edge over current leading methods in creating more accurate visual representations from text.

Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context

Gemini 1.5 Pro is a step up in the Gemini series, excels in multimodal understanding, juggling millions of tokens across text, video, and audio. This powerhouse nails near-perfect recall on retrieving detailed information, sets new records in QA tasks, and outperforms predecessors with its ability to handle up to 10M tokens—way beyond older models. It even picks up rare languages like Kalamang just from a grammar guide, showing the vast potential of large language models.

DeepSeek-VL: Towards Real-World Vision-Language Understanding

DeepSeek-VL presents an open-source model aiming at real-world understanding of both vision and language. It focuses on handling diverse and scalable data, including web screenshots and PDFs, to better reflect practical scenarios. The model also employs a hybrid vision encoder for efficiently processing high-resolution images while keeping computational demands low. With a strategy that balances vision and language learning from the start, DeepSeek-VL delivers top-notch performance in visual-language tasks and maintains strong language abilities, making it a standout choice for vision-language chatbots and other applications.

VideoElevator: Elevating Video Generation Quality with Versatile Text-to-Image Diffusion Models

VideoElevator boosts video generation quality by leveraging text-to-image diffusion models for sharper frames and better text alignment. It sidesteps traditional training hurdles by enhancing frame consistency and detail through a novel approach, mixing temporal motion refinement with spatial quality upgrades. This method significantly improves base text-to-video models, allowing for more realistic and stylistically varied video synthesis. The approach showcases a practical leap forward, especially in creating videos that match textual descriptions closely.

CogView3: Finer and Faster Text-to-Image Generation via Relay Diffusion

CogView3 is a groundbreaking model enhancing text-to-image generation with relay diffusion, a novel approach that first crafts low-res images then sharpens them up. This two-step process not only nails the details but slashes training and guessing time in half compared to the top dog, SDXL. It's faster, finer, and a real game-changer, making high-quality image creation more efficient and accessible.

🛠️TOP TOOLS

Tangle - summarize the best arguments from across the political spectrum on the news of the day

Wondercraft - Create ads, podcasts, audiobooks, meditations, ads and effortlessly translate your content for a global audience.

Balloonary - Build and launch online ads with the power of AI

Video to Blog - Create blogs from YouTube videos

HostAI - automate your ‍phone calls, ‍‍‍‍STR operation, ‍guest messages, open-night upsells and ‍maintenance ticketing.

Tusk - AI-created pull requests for annoying tickets

Type Prompt- Instantly generate human-like social posts with our AI-driven hook prompts

📲SOCIAL MEDIA

🗞️MORE NEWS

Reddit goes public: the latest updates on its IPO

Reddit is going public, aiming for a $6.5 billion valuation. After a tumultuous year, CEO Steve Huffman emphasizes the value of Reddit's data for AI training, securing a $60 million deal with Google. Despite past hiccups, Reddit offers shares to users and moderators in its IPO, with mixed expectations. THE VERGE

China Begins Testing AI Chatbot for Brain Surgeons in Hospitals

China is trialing an AI assistant for neurosurgeons in Beijing and other cities. Developed by a Hong Kong-based agency linked to the Chinese Academy of Sciences, the AI model utilizes Meta Platforms Inc.'s Llama 2.0. Trained on medical literature, it aims to support surgeons during procedures. BLOOMBERG

How the A.I. That Drives ChatGPT Will Move Into the Physical World

Covariant, a robotics startup, is pioneering AI technology that enables robots to learn physical skills akin to chatbots' digital capabilities. By integrating camera and sensory data with vast text datasets used for training chatbots like ChatGPT, Covariant's robots gain a deeper understanding of their surroundings, enhancing their operational abilities in warehouses and distribution centers. THE NEW YORK TIMES

UG Labs raises $7M to infuse conversational AI into games for kids

UG Labs, a Tel Aviv-based startup, secures $7 million to integrate conversational AI into children's games. Their technology, embedded in game engines, enables open-ended conversations with non-player characters, enhancing engagement and offering unique experiences. The AI recognizes children's speech patterns and fosters inclusivity by adapting to varying language abilities. VENTUREBEAT

Apple has begun testing an AI-powered ad product similar to Google's Performance Max as it looks to supercharge its $7 billion ad business

Apple is testing an AI-driven tool for App Store ads, similar to Google's and Meta's products. The tool automates ad placement, potentially expanding to other Apple platforms. Experts speculate on new ad spots in Maps, Apple TV+, and more. Apple's ad revenue is expected to hit $7 billion this year. BUSINESS INSIDER

Team behind popular Falcon AI models unveils new startup with $20 million in funding aimed at helping companies tailor LLMs for business

Adaptive, a startup founded by ex-Hugging Face and Falcon AI team, emerges with $20 million funding. They aim to simplify training large language models (LLMs) for businesses, focusing on reinforcement learning from human feedback (RLHF). Adaptive's platform aims to allow LLMs to learn from real interactions, offering solutions for ongoing training and fine-tuning. FORTUNE

Two AI-altered clips of Hitler’s 1939 Reichstag speech have racked up millions of views on X after being shared by a far-right troll.

The rise of AI-altered video clips featuring Adolf Hitler's 1939 Reichstag speech on X, a social media platform, has sparked controversy. Shared by a far-right influencer, these clips have garnered millions of views, prompting concerns about their antisemitic implications. While some viewers perceive the altered speech as favorable to Hitler's regime, others condemn its dissemination. WIRED

AI talent war heats up in Europe

Europe's AI talent war intensifies as startups flood the market, challenging DeepMind's dominance. High-profile exits like co-founder Mustafa Suleyman fuel competition. DeepMind offers senior researchers lucrative stock options to retain talent. Pay for AI executives skyrockets in Britain. Startups like Cohere and Anthropic lure talent with promising opportunities and competitive salaries. REUTERS

More precise understanding of dark energy achieved using AI

Researchers at University College London leverage AI to enhance understanding of dark energy's influence in the universe. Using AI, they double the precision of inferring key characteristics of dark energy from a map of dark and visible matter. This advancement helps rule out certain universe models and offers insights into dark energy's properties. PHYS ORG