MICROSOFT

🌐 Microsoft and OpenAI plan $100 Billion AI supercomputer

Microsoft, in partnership with OpenAI, is planning an ambitious project to build an artificial intelligence supercomputer named “Stargate”, with an estimated budget of $100 billion. The project, which is in the early planning stages, envisions creating a data center in the United States that will house millions of the latest AI chips. Stargate is expected to be operational by 2028.

The Stargate supercomputer is part of a series of five AI facilities that Microsoft and OpenAI intend to develop over the next six years as part of their long-term partnership. This massive investment signals the AI arms race among technology giants, with Microsoft likely funding most of the project. The partnership is evaluating potential chip suppliers, including Nvidia and AMD, and even considering self-sufficiency in hardware needs for Stargate. This development highlights the rapid escalation of the companies' AI ambitions and draws increasing regulatory scrutiny in the US, UK, and Europe over competition concerns.

Learn more:

1. businessinsider.com | 2. in.mashable.com

OPENAI

🎬 OpenAI in Hollywood: a revolution in filmmaking?

OpenAI is in talks with studios and Hollywood executives to introduce Sora, its AI tool capable of creating videos from text. Sam Altman, CEO of OpenAI, has been attending events in Los Angeles, aiming to bring filmmakers closer to technology by discussing partnerships for audiovisual content production. This move comes after the 2023 Hollywood writers' and actors' strike, which resulted in an agreement that included limitations on AI.

Sora promises to transform film production, offering a new way to create high-quality videos. However, it raises questions about copyright protection and regulation. It's essential to ensure that the references used by AI are transparent and that the talents and information of artists are not inappropriately appropriated. The entertainment industry may be on the verge of a significant change, but it's crucial that innovation occurs ethically and responsibly.

Learn more:

1. mashable.com

A16Z

🤖 The rise of generative AI in the business sector

The Andreessen Horowitz (a16z) report highlights a significant shift in how companies are building and acquiring generative AI. In 2023, generative AI conquered the consumer market, reaching over a billion dollars in spending. For 2024, the revenue opportunity is expected to be much greater in the business sector. Companies are almost tripling their budgets for generative AI, expanding the number of use cases implemented on smaller open-source models, and moving more workloads from initial experimentation to production.

Business leaders are reallocating AI investments to recurring software budget lines, indicating a long-term commitment to generative AI. A16z believes that AI startups that develop solutions for companies' strategic AI initiatives and anticipate their pain points will capture this new wave of investment and secure a significant market share. With increasing budgets and a more open attitude towards innovation, the business sector is positioning itself to take advantage of this disruptive technology.

Learn more:

1. a16z.com 

OPENAI

🎙️ OpenAI launches voice cloning tool

OpenAI introduced an innovative tool in the testing phase called “Voice Engine”, which allows developers to clone any voice with just a 15-second sample. The technology, in development for about two years, can generate a synthetic copy of the original voice from a small audio sample. However, the public availability date has not been announced yet, giving the company time to address the model's use and misuse.

The “Voice Engine” tool from OpenAI has the potential to transform how we interact with voice interfaces and can be applied across various sectors, from reading assistance to content translation. However, OpenAI acknowledges the potential for misuse, such as spreading misinformation or impersonating someone, and is working to ensure the technology is deployed responsibly. The company hopes to start a dialogue on the responsible use of synthetic voices and how society can adapt to these new capabilities.

Learn more:

1. techcrunch.com 

xAI

🤖 Grok 1.5: the next generation chatbot from xAI

xAI, Elon Musk's AI startup, announced the launch of Grok 1.5, a significant update to its chatbot model. Grok 1.5 promises notable improvements in reasoning capabilities, especially in coding and mathematical tasks. With a context extension of up to 128,000 tokens, the new model can process information from much longer documents, allowing for a deeper and more complex understanding of the prompts provided.

Grok 1.5 also stands out for its robust and flexible infrastructure, built on a distributed training framework based on JAX, Rust, and Kubernetes. This enables the xAI team to prototype ideas and train new architectures on a large scale with minimal effort. Moreover, the model will soon be available to early testers and existing platform users.

Learn more:

1. techcrunch.com 

INTEL

🖥️ Intel and Microsoft plan local AI PCs

Intel and Microsoft are collaborating to enable Microsoft's AI assistant service, Copilot, to run locally on PCs. This shift represents a significant advancement over the current cloud-based execution, which can introduce delays. Local execution of AI models can improve user privacy, allowing the use of AI software without automatically sending information for model training.

Furthermore, Intel confirmed that next-generation AI PCs would require a minimum performance of 40 trillion operations per second (TOPS) in the Neural Processing Unit (NPU). This is part of an "AI PC" project that will include an NPU, CPU, GPU, Microsoft's Copilot, and a physical Copilot key on the keyboard. PCs meeting these requirements are already being shipped, marking the beginning of the AI PC initiative. Qualcomm might be the first to offer an NPU suitable for powering Copilot locally, with its upcoming Snapdragon X chips promising up to 45 TOPS.

Learn more:

1. arstechnica.com | 2. tomshardware.com

HUME AI

🤗 Hume AI unveils Empathic Voice Interface

Hume AI raised $50 million in a Series B funding round and revealed the EVI (or “Empathic Voice Interface”), a technology reminiscent of the enchanting AI voice Samantha from the movie "Her". The EVI is powered by a new type of generative multimodal AI, called the Empathic Language Model (eLLM), and was trained with data from millions of human interactions.

EVI stands out for its ability to understand tone and volume to perceive when users have finished speaking, read their mood, and optimize responses over time for "happiness and satisfaction". The interface is not intended to simulate a relationship, but many users will likely feel comfortable sharing their deepest feelings and secrets with a convincing and emotionally attuned AI. A demonstration version of EVI is already active, and the company plans to make it publicly available to developers in April.

Learn more:

1. the-decoder.com | 2. demo.hume.ai

RESEARCH

🗣️ The persuasiveness of language models

The recent study published on arXiv, titled “On the Conversational Persuasiveness of Large Language Models: A Randomized Controlled Trial”, explores the persuasiveness of large language models in direct conversations with humans. The research, conducted by Francesco Salvi and collaborators, created a web platform where participants debated in multiple rounds with a live opponent. Participants were randomly assigned to one of four treatment conditions, which included debates between two humans or between a human and a language model, with or without access to basic sociodemographic information of the opponent.

The results showed that participants who debated with GPT-4, having access to their personal information, were 81.7% more likely to increase agreement with their opponents compared to those who debated with humans. Without personalization, GPT-4 still outperformed humans, but the effect was smaller and statistically insignificant. The study raises significant concerns about personalization and its implications for social media governance and the design of new online environments.

Learn more:

1. arxiv.org

• 🤖 LimX Dynamics introduced its bipedal robot P1, demonstrating great control and balance as it autonomously navigated an unmapped forest. Learn more.

• 🧬 Insilico Medicine achieved a significant milestone by starting Phase 2 clinical trials for INS018_055, the first drug discovered and designed by AI to treat fibrosis. This breakthrough in AI-driven drug discovery could significantly accelerate the development of new treatments for diseases like idiopathic pulmonary fibrosis (IPF), affecting millions worldwide. Learn more.

• 🎖️ The National WWII Museum in New Orleans opened an exhibition that allows visitors to converse with AI-driven video images of veterans and Holocaust survivors. The exhibition, called “Voices from the Front”, uses artificial intelligence and voice recognition software to index memories so visitors can "talk" with Americans from the WWII era for decades to come. Learn more.

• 🎨 Adobe unveiled significant innovations at the Adobe Summit, highlighting GenStudio and Firefly, its new AI tools for customer experience management (CXM). GenStudio is an AI platform for marketing teams, enabling them to plan, create, manage, and activate brand content quickly. Learn more.

• 🇪🇺 The European Union issued a warning about the emergence of deepfakes in the elections of several member states, raising concerns about the dangers of artificial intelligence and possible interferences. The European Commission, led by Vice-President Vera Jourova, released guidelines for major digital platforms on how to protect against online risks to elections, including generative AI. Learn more.

• 🎵 Tennessee became the first US state to pass a law protecting musicians and other artists against the misuse of AI. The legislation, known as the “ELVIS Act”, aims to ensure that AI tools do not replicate an artist's voice without their consent. The measure, effective from July 1st, creates a new civil action where individuals can be held liable for publishing or performing someone's voice without permission, as well as using technology to produce the name, photographs, voice, or likeness of an artist without proper authorization. Learn more.

• 👩‍⚕️ NVIDIA, in collaboration with Hippocratic AI, announced a partnership to offer virtual nurses powered by generative artificial intelligence. These AI nurses, costing only $9 per hour, aim to provide consistent and reliable healthcare services, operating 24/7 and helping to mitigate the widespread staffing shortage and increase access to high-quality care. Learn more.

• 🌐 Cooperation between the United States and Japan in artificial intelligence and semiconductors is set to intensify. A joint statement, to be announced during Prime Minister Fumio Kishida's official visit to the US in April, will emphasize global partnership and stronger collaboration in these high-tech areas. Learn more.

• 🍺 Researchers at KU Leuven University in Belgium used artificial intelligence to enhance beer flavor. The team led by Prof. Kevin Verstrepen built machine learning models to predict the taste and appreciation of beer based on its composition, analyzing over 200 compounds involved in flavor. Learn more.

The video of the week is actually a series of videos: OpenAI published an article with the first tests of its Sora model by some selected artists — and the results are impressive. Click here or on the image above to watch.

• 👩🏽‍🏫 Gatekeep: an AI platform that transforms questions into educational explanatory and personalized videos. Ideal for those looking to learn more quickly and efficiently, Gatekeep creates AI videos that make complex concepts, such as mathematics, more accessible and easier to understand. Link.

• 🎨 RenderNet: an AI image generation tool, focused on professional users. Based on the stable diffusion model, the platform offers over 100 models to create sophisticated images quickly. With smart features like Facelock and multi-model image generation, RenderNet facilitates the production of complex images in seconds. Link.

• 🗣️ Speak: a versatile platform offering automatic transcription services, professional transcription, and AI-based solutions, such as a meeting assistant and generative artificial intelligence. It is ideal for industries like marketing and research, or any team that requires data visualization and linguistic data analysis. Link.

• 🤝 Groupthink: an AI tool that combines ChatGPT, Slack, and a web search engine to enhance team collaboration. The platform allows teams to add AI analysts to their conversations for real-time business insights and web searches. Link.

🔎 Access Google Gemini 1.5 for Free

In this tutorial, you'll learn how to access Google Gemini 1.5 for free and take advantage of its context window of up to 1M tokens — enough to upload large files or even long videos. Step by step:

1. Visit the Google DeepMind website.

2. Click on “Gemini 1.5” or scroll down to where it says “Introducing Gemini 1.5”

   

3. Next, press “Try Gemini 1.5” and log in with your Google account.

4. Done! You should see a screen that takes you directly to Google's Gemini 1.5 chatbot. The 1.5 model has the capacity to analyze large files, allowing you to “converse” with PDFs, photos, videos, and other documents.