OpenAI Unveils Sora

PLUS: Salesforce $1 Billion Investment, Lambda’s $1.5 Billion Valuation and more.

Today:

OpenAI Unveils A.I. That Instantly Generates Eye-Popping Videos

OpenAI debuts Sora, an A.I. tech producing lifelike videos, following Runway AI's initial blurry attempts. The system, likened to Hollywood-grade footage, raises concerns about online disinformation. Named after the Japanese word for sky, Sora is shared with a select group for testing due to potential misuse risks. Watermarking helps identify A.I.-generated content, but removal is possible. 

Sora utilizes generative A.I., learning from video data and captions. Despite impressive results, occasional flaws occur, like a cookie that never shrinks. The rapid advancement of A.I. image generators like DALL-E poses challenges for digital artists, leading to job losses.

Meet Gemini 1.5, Google's newest AI model with major upgrades from its predecessor

Google unveils Gemini 1.5, an upgraded AI model with significant improvements over its predecessor. Despite Gemini 1.0's recent launch, the new model boasts enhanced performance, including a longer context window and better understanding. CEO Sundar Pichai praises its capabilities, claiming comparable quality to the advanced 1.0 Ultra model while using less compute. 

Built on a new MoE architecture, Gemini 1.5 can process up to one million tokens, significantly expanding its capacity. Google offers a limited preview to developers and enterprise customers for testing, emphasizing safety and responsible deployment. Pricing tiers will vary based on token context window size.

V-JEPA: The next step toward Yann LeCun’s vision of advanced machine intelligence (AMI)

Meta AI introduces V-JEPA, a model advancing machine intelligence with a more realistic understanding of the world. V-JEPA excels at detecting detailed object interactions in videos, trained through self-supervised learning without labeled data. By predicting missing video parts in an abstract space, it achieves efficiency and flexibility, outperforming previous models in various tasks. 

Unlike prior methods requiring full fine-tuning, V-JEPA allows reuse of pre-trained parts for different tasks, promising versatility. Future research aims to incorporate audio and extend V-JEPA's capabilities for longer time predictions, paving the way toward Advanced Machine Intelligence (AMI). The model is released under a Creative Commons license, fostering open science.

OpenAI Develops Web Search Product in Challenge to Google

OpenAI, led by Sam Altman, is challenging Google with a new web search tool, leveraging Bing's technology. This move intensifies competition in the search engine market. The product aims to offer an alternative to Google, potentially disrupting its dominance. OpenAI's initiative reflects a broader trend of tech startups challenging industry giants. The project's success could depend on factors like user adoption and the quality of search results. 

This development signifies OpenAI's ambition to expand its influence beyond AI research and into consumer-facing products. With backing from Microsoft, OpenAI's venture poses a formidable challenge to Google's search hegemony.

Lambda Hits $1.5 Billion Valuation for AI Computing

Lambda, a startup specializing in cloud computing for AI, secured $320 million in funding, boosting its valuation to $1.5 billion. Led by Thomas Tull's US Innovative Technology, the investment includes backers like B Capital and SK Telecom. 

Existing supporters, including Crescent Cove and Bloomberg Beta, also contributed. The funding underscores growing interest in AI infrastructure.

Salesforce to Invest in Nvidia Chip Reseller at More Than $1 Billion Valuation

Salesforce is set to invest over $1 billion in Together AI, a reseller of Nvidia's AI chips. The surge in demand for Nvidia's AI chips has boosted companies like Together AI. The startup aids AI developers in accessing Nvidia server chips for training and refining AI models. 

This new funding round, doubling Together AI's valuation to over $1 billion in just four months, will be led by Salesforce Ventures, with Coatue Management also participating. The move underscores the growing importance of AI technology and the significant investments being made in its development.

Germany: Microsoft to invest €3.3 billion in AI capacities

Microsoft is dropping a whopping $3.5 billion in Germany to beef up its AI and cloud computing game, marking its biggest investment there in four decades. The plan? To double down on their tech infrastructure and get a million-plus Germans savvy in AI skills. Chancellor Olaf Scholz is all in, seeing it as a big thumbs-up for Germany's open-market vibes and a boost to keep the country on the tech forefront. 

This move is especially big news for North Rhine-Westphalia, getting the lion's share of the investment. It's a clear sign Microsoft is betting big on Germany, aiming to close the skills gap and tighten ties with major local players like Bayer and RWE.

🧠RESEARCH

"Magic-Me" is a framework for creating videos that maintain a person's unique identity throughout. By using a few images to define the subject's ID, it ensures that the person's likeness is consistently recognizable in every frame of the video. This system combines new techniques for separating the subject from background noise and improving visual consistency and clarity. It's a big step forward for making personalized videos that look sharp and stay true to the person's appearance. 

The paper highlights how the order in which premises are presented affects the performance of large language models (LLMs) in reasoning tasks. Surprisingly, when premises are arranged in a sequence that mirrors the steps needed for solving a problem, LLMs do better. This insight came from observing a significant drop in performance—over 30%—when the order was shuffled. The study introduces a new benchmark, R-GSM, to further explore this phenomenon in mathematical problem-solving. This discovery points to a critical aspect of LLM training and optimization for enhanced reasoning capabilities. 

L3GO is an innovative system for creating 3D models of unconventional objects through a process that involves large language models acting as agents within a 3D simulation. Unlike traditional models that struggle with out-of-the-ordinary designs, L3GO excels by using trial-and-error to assemble parts into a coherent whole. It's tested against a new benchmark, showing superior performance in generating 3D shapes that defy standard expectations. This approach marks a significant advancement in the field of 3D object generation.

"GhostWriter" is about enhancing AI-assisted writing by making it more personalized and user-friendly. The tool learns a user's writing style on-the-fly and allows for manual tweaks to ensure the text sounds just right. Testing with 18 people showed it made writing with AI more tailored and controlled, leading to better collaboration between humans and AI in crafting texts. This approach could shape the future of writing aids, offering key insights for design improvements.

🛠️TOP TOOLS

Sora = AI model that can create realistic and imaginative scenes from text instructions.

Sheet Copilot -  uses AI to tackle tasks and answer questions in Google Sheets

Instant Summaries -  AI feature that provides quick, smart summaries at the top of your email threads, designed to save time and reduce inbox stress.

Shopia - a platform designed to streamline content creation using AI, offering tools for writing, researching, and automating tasks efficiently. 

Shortform - offers in-depth guides and summaries on a wide range of non-fiction books and articles. 

Deformity - online forms with AI-driven designs that stand out, aiming to transform the traditional form experience into something more engaging and effective.

📲SOCIAL MEDIA

🗞️MORE NEWS

Largest text-to-speech AI model yet shows ’emergent abilities’

Amazon researchers unveil the largest text-to-speech AI model, boasting improved natural speech abilities, potentially bridging the uncanny valley. Dubbed BASE TTS, the model demonstrates emergent behaviors, handling complex sentences with finesse. Despite its experimental nature, the breakthrough holds promise for enhanced conversational AI, notably in accessibility and stream ability. TECHCRUNCH

Apple buys domain suggesting generative AI additions may come to iWork

Apple has acquired the domain "iWork.ai," sparking speculation about potential AI features in its iWork suite. While some suggest a relaunch to compete with Microsoft Office and Google Docs, others caution it could be a move to protect intellectual property or prevent domain misuse. Expectations for AI integration in iOS at WWDC 2024 persist. APPLE INSIDER

Open Source AI Software Maker LangChain Launches First Paid Product — With A Massive Waitlist

LangChain, led by CEO Harrison Chase, launches LangSmith, aiding developers using large language models like OpenAI's GPT-4 to test and track their code. With a $20 million funding round led by Sequoia, LangChain's LangSmith boasts an 80,000-person waitlist, attracting notable companies like Elastic and Rakuten. LangChain's journey from open-source project to high-demand AI tool reflects the rapid growth of the AI industry. FORBES

Eye of Sauron for methane: Google, green group launch "unprecedented" tracker

Google and Environmental Defense Fund team up to track methane emissions with satellite technology and AI, likened to "Eye of Sauron." Methane, a potent greenhouse gas from oil and gas, poses significant climate risks. The initiative aids corporate and regulatory efforts, aligning with global pledges to cut methane emissions. AXIOS

Stability AI tries to stay ahead of the pack with a new image-generating AI model

Stability AI introduces Stable Cascade, a more efficient and powerful image-generating AI model than its predecessor, Stable Diffusion. It offers various editing features like resolution enhancement and inpainting. The model, available for research on GitHub, utilizes a novel architecture for faster processing. Stability AI faces legal challenges over copyright infringement allegations. Commercial licenses are now offered for funding research. THE VERGE

These lab-grown hearts made from human stem cells are used to train AI to detect heart diseases

Lab-grown hearts made from human stem cells are being used to train AI in detecting heart diseases faster. These "organoids" are destroyed after 28 days. The aim is to develop AI software recognizing detailed heart movements in medical images. Researchers also focus on removing harmful old heart cells using AI. EURONEWS

AI startup Flower Labs, whose software allows AI models to be trained ‘at the edge,’ is valued at $100 million in new funding round

Flower Labs, a startup pioneering software enabling AI training at the edge, secures a $100 million valuation in a recent funding round led by Felicis Ventures. Their technology facilitates federated learning, crucial for industries like healthcare and finance. The founders, originating from the University of Cambridge, aim to expand their platform's reach with the investment. FORTUNE

Stanley Druckenmiller exits Alphabet, keeps winning AI plays Nvidia and Microsoft

Billionaire investor Stanley Druckenmiller sells off Alphabet shares but sticks with winners like Nvidia and Microsoft. Druckenmiller's move signals confidence in AI-focused investments, reflecting ongoing shifts in tech holdings among top investors. This shift suggests a bullish sentiment towards AI plays while reducing exposure to certain tech giants. CNBC

UAE Backs Sam Altman Idea to Turn Itself into AI Testing Ground

The UAE aims to lead in AI testing and regulation, inspired by OpenAI CEO Sam Altman's vision. Minister of State for Artificial Intelligence, Omar Al Olama, sees this as a chance for the UAE to be a major player in AI. The country plans to advocate for becoming a hub for practical AI regulation. BLOOMBERG

Get business/tech news just with graphics

We explain the latest business, finance, and tech news with visuals and data. 📊

All in one free newsletter that takes < 5 minutes to read. 🗞

Save time and become more informed today.👇

What'd you think of today's edition?

Login or Subscribe to participate in polls.

What are MOST interested in learning about AI?

What stories or resources will be most interesting for you to hear about?

Login or Subscribe to participate in polls.

Reply

or to participate.