FlagThis AI

This cluster covers research in quantum computing and cryptography. It examines the development of quantum-resistant cryptography and quantum software and also highlights the integration of quantum computing systems in global markets. It explores the practical applications and theoretical foundations of quantum technologies.

Microsoft is suing a foreign-based hacking group for exploiting Azure AI services to generate harmful content. The group operates a hacking-as-a-service infrastructure, intentionally bypassing safety controls of Microsoft’s generative AI services. They used stolen Azure API keys and customer Entra ID credentials. They then sold tools to other malicious actors who used them to produce offensive and harmful content. Microsoft is taking legal action to stop this misuse of their AI technology.

Mistral has released an upgraded version of their code generation model, Codestral 25.01, which is faster and more efficient than its predecessor. It demonstrates a 256k token context, which is a new record for Mistral. This model is designed for code completion and is making significant strides in the developer community as a powerful coding assistant. This new version scored impressive first place on the Copilot Arena leaderboard and 11% on the aider polyglot benchmark. This code focused model is not available as open weights but is accessible through an API.

Nvidia is collaborating with multiple healthcare and biotech companies like Illumina and Mayo Clinic to advance drug discovery and genomic research. The collaboration combines Illumina’s sequencing technologies with NVIDIA tools to develop biological foundation models and accelerate the analysis of multiomic data. Additionally, Mayo Clinic is expanding the use of AI-powered clinical documentation, leveraging tools from Abridge to analyze and interpret data which could transform pathology and accelerate medical advancements.

GitHub has launched Copilot Workspace, an AI-powered coding environment that allows users to define tasks in natural language and collaborate with AI. This tool is designed to enhance developer productivity by assisting with coding tasks. The tool is now generally available without a waitlist. This is another major step forward for AI enabled coding. Github also introduced the capability to write code in native languages such as Kannada.

Salesforce has released Agentforce for Retail, an AI agent skills library specifically for retailers, allowing them to create AI agents that support customer service workers, store associates, and customers. Salesforce has indicated that they are not hiring more software engineers because their AI tools have already increased productivity by over 30%. SharkNinja is now using Agentforce to power their customer service.

The rise of AI-powered traffic cameras and smart city solutions for advanced traffic monitoring is under discussion. These systems use AI to analyze traffic data, optimize traffic flow, and enhance road safety. Concerns have been raised about the privacy of collected data. Smart cameras provide advanced adaptive traffic monitoring, which helps in effective urban traffic management.

Several companies are leveraging AI to enhance Retrieval-Augmented Generation (RAG) applications. Contextual AI is specializing in RAG agents for expert knowledge work, while new tools like Ragie aim to simplify building AI applications with RAG. Other tools like IntelliBricks also aim to provide a streamlined way to use RAG. The industry is clearly pushing RAG as a key technology to augment AI capabilities and make them more useful and functional.

François Chollet, the creator of Keras and the ARC benchmark, has launched Ndea, an AI research lab, focused on creating AI that learns like humans. Ndea aims to develop more robust AI systems by focusing on deep learning guided program synthesis, and the team is actively hiring remote researchers. The core focus is on building more general AI that can improve itself. This is an alternative approach to NVIDIA’s hardware acceleration approach.

Meta has developed an advanced AI model called SeamlessM4T, capable of translating speech and text across 101 languages, including direct speech-to-speech translation for 36 languages. This model, trained on millions of hours of multilingual audio, significantly improves upon existing systems in both accuracy and noise resilience. It also provides multiple supporting AI systems which can be used for analyzing speech and text which have other use-cases including detecting early signs of cognitive decline using voice patterns. Meta is now utilizing the technology to enable real time translation in many of its products including RayBan Meta and Instagram.

Various researchers have published articles explaining concepts and methods in statistics, probability, and machine learning. There seems to be a surge in the amount of educational material being made available to data professionals which is leading to a rise in AI/ML understanding of the community. This indicates a growing need for professionals to have a deeper understanding of these topics.

AI is being used in various research projects. A new agentic open source framework named ‘Agent Laboratory’ has been released by Johns Hopkins University and AMD which is aimed at accelerating scientific research. A new technique called Content-Adaptive Tokenizer (CAT) has been released which adapts token count based on image complexity. Renmin University of China has developed a Search-o1 Agentic Search-Enhanced model, while also an open-source dataset for math problem solving has been released. These all point to a steady stream of innovative AI research and development.

OpenAI’s ChatGPT is enhancing its capabilities by introducing a new ‘Tasks’ feature, which brings it closer to becoming a comprehensive AI assistant. This feature allows users to schedule reminders and recurring requests, automating daily workflows. This enhancement positions ChatGPT as a more versatile tool capable of handling routine tasks, moving beyond its text-based interactions. Users can now instruct the AI to perform scheduled actions, such as daily news reports or sports scores. The feature is still in Beta but is a significant step towards making ChatGPT a full-fledged AI assistant.

OpenAI is working with media companies to integrate AI into journalism. Axios is partnering with OpenAI to distribute their content through ChatGPT and also get access to OpenAI’s technology. This includes summaries, quotes, and links. This is a new approach from news organization to use AI to expand and improve the distribution and discoverability of their content. It also highlights the financial challenges of journalism in the age of AI, as these companies look to new ways to generate revenue.

IonQ is a leader in quantum computing, showing potential for transforming machine learning and AI. Despite a recent stock sell-off, the company has achieved milestones including a new manufacturing facility and delivery of its Forte system. While there is skepticism from industry leaders regarding the timeline for practical applications, IonQ is projected to experience significant revenue growth, positioning itself in a market projected to be $65 billion by 2030. They have also partnered with Amazon to provide cloud services for quantum computing.

Adobe is offering an AI-powered feature called ‘Bulk Create’ in Firefly, which allows users to edit multiple images at once. This is designed to help enterprise creatives who need to frequently re-edit the same images. However, some users have had negative experiences with Adobe’s generative AI features, including cases where the AI has produced unexpected and unwanted results such as auto-filling a seagull photo with a bitcoin, which indicates issues with the quality and reliability of the image generation. There also seems to be some confusion over changes to their 20GB Photoshop/Lightroom plan, which is no longer available to new customers after January 15, 2025.

NVIDIA’s new AI model, Sana, enables free 4K image generation on local laptops. Nvidia is also working on AI-powered CPUs running Linux, marking a significant step in hardware technology for AI. Nvidia has also begun full-scale production of the Blackwell GPU for AI and HPC, with Blackwell revenue expected to surpass Hopper’s in early 2025.

Quantum computing stocks experienced a significant collapse due to remarks made by Nvidia CEO Jensen Huang, who suggested that truly useful quantum computers are still 15 to 30 years away. This statement caused a major correction in the valuations of companies like IonQ, Quantum Computing Inc, and D-Wave. Despite this, IonQ CEO maintains a target of $1 billion in revenue by 2030. The industry faces challenges with commercialization, as the tech is still in the early stages of development. NVIDIA and other major tech companies are investing billions in quantum technologies, but it’s still uncertain how or when the business will become profitable, and these companies will be the leaders. Despite the high market caps, the total combined revenue is under 50 million.

Google is testing ‘Daily Listen’, an AI feature that creates personalized podcast-style summaries using user’s search history and news preferences in the Discover feed, similar to NotebookLM’s Audio Overviews but on a regular basis. This feature highlights how Google is using AI to enhance personalized content delivery. Additionally, Google is improving AI for developers with Grounding with Google Search, aiming to reduce non-determinism in AI models. Google Cloud is also expanding, with a new region in Thailand, reflecting the company’s push in AI and Cloud.

This cluster discusses Quantum computing and its mathematical applications. Topics include in-depth exploration of quantum bits using Qiskit, and mathematical representations of quantum phenomena. Also, the cluster includes an article on the nature of quantum mechanics and its counterintuitive behavior. These articles explain the mathematical concepts which are essential for quantum computing.