• AI Breakfast
  • Posts
  • The Boundless Potential of Quantum Computing

The Boundless Potential of Quantum Computing

Author

AI Breakfast
July 12, 2023

Good morning. It’s Wednesday, July 12th.

In today’s email:

  • The Boundless Potential of Quantum Computing

  • TCS boosts Azure Open AI training; new AI service.

  • AMD CEO: AI to lead chip design.

  • AI security cameras for VIPs tested in Japan.

  • GPT-4: 1.8 trillion parameters, high costs leaked.

  • AI eyed for women's health solutions.

  • Chip makers use 'chiplet' stacking for AI.

  • Character.AI battles waifu, copyright issues.

  • Pano AI gets extra $17M for wildfire detection.

  • 5 New AI Tools

  • Latest AI Research Papers

You read. We listen. Let us know what you think of this edition by replying to this email, or DM us on Twitter.

Today’s edition is brought to you by:

Raindrop helps you design personalized trading strategies using AI.

Whether you're a beginner or an experienced trader, Raindrop helps you write custom trading bots tailored to your investment strategy.

Simplify and automate your portfolio management with Raindrop’s AI Bot Marketplace.

Can you program an intelligent trading bot of your own?

Today’s trending AI news stories

The Boundless Potential of Quantum Computing

Google’s 70-Qubit Quantum Computer

The journey to harness the full potential of quantum computing has only just begun.

Google's scientific team recently announced a significant milestone in quantum computing, successfully completing a computational task that would take a traditional supercomputer 47 years to complete.

The task, a random circuit sampling calculation, was executed on an updated version of the Sycamore processor, now with a capability of 70 qubits, and the accomplishment has been considered another assertion of quantum advantage.

A qubit is the basic quantum information unit, equivalent to a bit in classical computing. The addition of more qubits to the Sycamore processor has reportedly increased its power by 241 million times compared to the previous 53-qubit model, as the computational capability exponentially increases with the number of cubits.

The scientists specifically designed the experiment to address lingering queries about the comparative capabilities of classical and quantum computers. They aimed to define boundaries for the utility of the exponentially large Hilbert space in a noisy quantum processor and to identify an experimental observable probing these boundaries.

Prominent voices in the field, like Steve Brierley, CEO of Cambridge-based quantum company Riverlane, hailed the achievement as a major milestone, putting to rest debates about the achievement of quantum supremacy. Conversely, Sebastian Weidt, CEO of Brighton-based start-up Universal Quantum, emphasized the need for quantum computers to focus on practical functionalities.

While the experiment's current practical implications may be [very] limited, it's hoped that such advanced research can pave the way for quantum computing applications that are more practically useful.

Quick News

TCS Bets Big on Azure Open AI: Plans to Train 25,000 Engineers; Launches New Generative AI Offering for Clients: Tata Consulting Services (TCS) has announced its plan to scale its Azure OpenAI expertise and launch its Generative AI Enterprise Adoption offering on Microsoft Cloud. TCS aims to train 25,000 engineers on Azure Open AI to help clients accelerate their adoption of generative AI. The Generative AI Enterprise Adoption offering combines TCS’ contextual knowledge with Azure Open AI to enhance customer experience, launch new business models, and improve productivity.

CEO of AMD: AI Will Dominate Chip Design: Lisa Su, CEO of AMD, predicts that AI will dominate chip design as the complexity of processors increases. AMD already uses AI in chip design, testing, and verification, and plans to leverage generative AI more broadly in future applications. AI is also being used by leading electronic design automation (EDA) toolmakers, such as Ansys, Cadence, and Synopsys, to support chip designs. Synopsys launched Synopsys.ai, the first end-to-end AI-driven EDA solution, allowing developers to utilize AI throughout all stages of chip development.

Japan police to test AI-equipped cameras in protecting VIPs: Japanese police will test security cameras equipped with AI to protect high-profile public figures. This move comes as Japan commemorates the anniversary of the assassination of the former Prime Minister Shinzo Abe. The AI-based cameras are expected to enhance existing security measures by detecting suspicious activity. The National Police Agency plans to commence testing within the current fiscal year, which ends in March 2024.

GPT-4's parameter and training details leaked (Reddit): GPT-4 is reported to have over 1.8 trillion parameters across 120 layers and utilizes a mixture of experts (MoE) model. It is trained on approximately 13 trillion tokens and has a 32k sequence length. The training cost for GPT-4 is estimated to be around $63 million, and inference costs are higher compared to previous models. Other details include parallelism strategies, dataset mixture, and speculation about its performance.

AI may be key to solving the most neglected women's health issues: Researchers are using AI and computational medicine to analyze large data sets and extract insights into conditions such as maternal mortality, breast cancer, endometriosis, and more. The combination of AI’s machine learning capabilities and personalized approaches based on individual patient data is expected to lead to improved care for women’s health concerns. The affordability and speed of computer runs have facilitated these advancements, and the increasing involvement of young female engineers, has also contributed to progress in the field.

To Drive AI, Chip Makers Stack ‘Chiplets’ Like Lego Blocks: Chip makers like Nvidia, Intel, and AMD are investing in technology that allows them to stack chips together like Lego blocks to accelerate the development of AI designs. This innovative approach promises more powerful and easier-to-build semiconductors, catering to the growing demand for AI-driven technologies. As the AI boom continues, chip stacking emerges as a key strategy in advancing AI capabilities.

Unicorn startup Character.AI struggles with waifus and copyrights: Character.AI, a unicorn startup founded by former Google engineers, is facing challenges related to waifus (characters from erotic manga and anime) and copyright issues: The platform, hosting 16 million chatbots and attracting over 200 million monthly visits, allows more users to create chatbots resembling popular characters and engage in romantic role-playing. However, explicit content and copyright infringement concerns have arisen. Character.AI sees itself as a democratizing force in AI technology and recently launched a subscription service offering perks to users.

Wildfire detection startup Pano AI extends its $20M Series A with another $17M: Wildfire detection startup Pano AI has raised an additional $17 million in a Series A extension, bringing its total Series A funding to $37 million. The company uses high-definition remote-controllable cameras mounted in strategic locations to detect fires and issue early warnings. Its AI algorithms analyze video footage for signs of smoke and fire, allowing emergency responders to take prompt action.

🎧 Did you know AI Breakfast has a podcast read by a human? Join AI Breakfast team member Luke (an actual AI researcher!) as he breaks down the week’s AI news, tools, and research: Listen here

5 new AI-powered tools from around the web

Meet for Slack simplifies the process of starting and sharing Google Meet calls directly within Slack. Connect with your team instantly, schedule meetings effortlessly, and sync your status and calendar automatically. Built by creators of Trivia, trusted by over 40,000 organizations worldwide.

Userdesk offers a no-code solution to automate customer support with AI ChatBots trained on your own data. Connect sources like Notion website, PDFs, and more to create a ChatGPT-like Chatbot in minutes. Reduce support volume, save time, and increase customer satisfaction.

Opus Clip is a generative AI video tool that converts long videos into short, viral clips with a single click. The AI analyzes the video, extracts highlights, and rearranges them into engaging short videos. Opus Clip provides additional features like active speaker detection, AI keyword highlighter, and AI emoji generator to enhance the clips.

BlazeAI is an advanced acquisition platform that utilizes AI to unlock business growth through social platforms. With a track record of helping over 100 companies increase revenue, Blaze enables targeted audience connection and impactful messaging.

Swimm AI is a powerful code documentation tool that streamlines the process of creating and maintaining documentation. It generates and suggests document structures, adds code explanations, and enhances document visibility.

arXiv is a free online library where scientists share their research papers before they are published. Here are the top AI papers for today.

VampNet is a masked acoustic token modeling approach for music synthesis, compression, inpainting, and variation. It leverages a variable masking schedule and a bidirectional transformer architecture to generate coherent music waveforms. VampNet can be prompted in various ways to perform tasks like music compression, inpainting, continuation, and looping with variation. It maintains style, genre, and instrumentation, making it a powerful music co-creation tool. Code and audio samples are available here.

OptiGuide is a framework that utilizes LLMs to provide insights and explanations for supply chain optimization outcomes. OptiGuide integrates LLMs with optimization solvers, allowing users to input queries in plain text and receive understandable insights about optimization results. The framework aims to enhance decision-making processes, enable what-if analysis, and facilitate communication between operators, program managers, and data scientists. The paper also introduces an evaluation benchmark for LLM accuracy in supply chain scenarios.

Semantic-SAM is a universal image segmentation model that enables segmenting and recognizing objects at any desired granularity. It offers semantic awareness and granularity abundance by consolidating multiple datasets and training on decoupled object and part classification. The model uses a multi-choice learning scheme to generate masks at multiple levels based on user clicks. It combines various segmentation tasks and achieves performance improvements. By jointly training on different datasets, Semantic-SAM achieves strong performance in multi-level segmentation.

The paper explores the role of international institutions in governing advanced AI systems. It suggests that international collaborations can unlock the benefits of AI for sustainable development, while coordinated regulatory efforts can reduce obstacles to innovation. The paper identifies four institutional models that we can build from: a Commission on Frontier AI, an Advanced AI Governance Organization, a Frontier AI Collaborative, and an AI Safety Project. These models aim to address challenges such as testing international standards, promoting access to advanced AI, and conducting quality AI safety research.

AnimateDiff is a framework that enable personalized text-to-image models that generate animated images without specific tuning. It introduces a motion modeling module into a base text-to-image model and fine-tunes it on video clips to learn motion priors. Once trained, the motion module can be inserted into personalized models, resulting in diverse and customized animations. AnimateDiff eliminates the need for data collection or customized training for each personalized model. The framework is evaluated on various personalized models and demonstrates the generation of pretty smooth animation clips while preserving domain and diversity.

Thank you for reading today’s edition.

Your feedback is valuable.


Respond to this email and tell us how you think we could add more value to this newsletter.

Attending Ai4 this year? We will be!