AI News

Brilliant Labs has launched Frame, the first AI glasses with an integrated multimodal AI assistant named Noa. Backed by notable figures in tech, it aims to transform daily interactions by combining stylish design with advanced AI capabilities. Pre-orders start March 2024 at $300.

Google's AI chatbot, Bard, has rebranded to Gemini, introducing a subscription-based Gemini Advanced mode. The update includes a dedicated Android app and enhanced integration into Google Workspace, aiming to outperform competitors like OpenAI. Gemini Advanced is available in over 150 countries, focusing on improved user experience and functionality.

Microsoft has redesigned its Copilot AI chatbot, enhancing its interface and introducing a carousel of suggested prompts. Promoting it through a Super Bowl commercial, Copilot has facilitated 5 billion chats and images. New features include advanced image editing and the upcoming Designer GPT for idea visualization.

Researchers from Google DeepMind and USC have introduced a "self-discover" prompting framework for large language models (LLMs) that enhances reasoning by up to 32% over existing methods while using significantly less computational power. This innovative approach enables LLMs to autonomously create task-specific reasoning structures, promising advancements towards general intelligence and effective Human-AI collaboration.

The Biden administration appointed Elizabeth Kelly as director of the new U.S. AI Safety Institute (USAISI) at NIST, focusing on AI risk management. Funding remains uncertain, with a bipartisan $10 million request pending, and concerns about transparency and operational details persist.

Kyndryl and Google Cloud are expanding their collaboration to develop responsible generative AI solutions. This partnership will integrate Google’s Gemini LLM with Kyndryl's expertise, enhance ERP data value, and introduce custom LLMOps. The initiative aims to facilitate business transformation through advanced AI technologies.

IBM is enhancing AI security through its open-source Adversarial Robustness Toolbox (ART), now available on Hugging Face. This initiative, developed since 2018 and supported by DARPA, aids developers in mitigating risks like data poisoning and evasion tactics, fostering robust AI systems.

Two Texas companies, Life Corporation and Lingo Telecom, are under investigation for robocalls using AI deepfakes that impersonated President Biden to deter New Hampshire voters. Up to 25,000 calls were made. The FCC has issued cease-and-desist letters, and new regulations on AI robocalls are forthcoming.

OpenAI has updated ChatGPT and DALL-E 3, adding metadata to identify AI-generated images based on C2PA standards. This follows Meta's similar labeling efforts. OpenAI’s metadata, which can be removed, aims to enhance transparency amid rising misinformation concerns, rolling out to mobile users by February 12.

Apple launched "MGIE," an open-source AI model for image editing using natural language instructions. It utilizes multimodal large language models (MLLMs) for precise modifications, enabling tasks from color adjustments to object manipulations. MGIE is user-friendly, accessible on GitHub, and enhances creative processes across various applications.

Cimba.AI, a new AI startup founded by ex-Airbnb and AWS professionals, raised $1.25 million to create a platform for enterprises to build custom AI agents. Targeting non-technical users, it streamlines workflows and automates tasks, aiming to enhance business operations significantly.

Microsoft has launched Face Check, a facial recognition feature in its Entra Verified ID platform, enabling businesses to verify identities by matching live selfies to ID documents. It enhances security, prioritizes privacy, and uses Azure AI services. Early adopters like BEMO have implemented it successfully, as Microsoft focuses on integration and open standards in the identity verification market. Face Check is currently in free preview.

Update: Tesla's "We, Robot" Event Wrap-Up

Researchers are enhancing the interpretability of large language models (LLMs) through the JumpReLU sparse autoencoder (SAE). This approach improves feature tracking in LLMs, balancing sparsity and reconstruction fidelity, while effectively minimizing irrelevant features and promoting understanding of LLM behavior, including bias mitigation.

AI interviews are gaining popularity in recruitment, streamlining initial screenings and reducing costs. However, they may overlook interpersonal skills and pose risks like data security and algorithmic bias. A balanced approach combining AI with traditional methods is recommended for effective hiring.