NerdNews

Groq, an AI chip startup and Nvidia challenger, is nearing a new fundraising round of $600 million at a $6 billion valuation. This doubles their previous valuation of $2.8 billion in August 2024. The round is led by Austin-based firm Disruptive.

Google's NotebookLM now offers Video Overviews, a feature that turns dense multimedia into digestible visual presentations. It's a visual alternative to Audio Overviews, allowing users to customize their experience and focus on specific topics or learning goals.

Google's AI Mode gets new 'Canvas' feature, real-time help with Search Live, and more. The Canvas feature helps users build study plans and organize information, while Search Live provides real-time assistance with visual context from the camera feed.

Microsoft is in advanced talks with OpenAI to secure ongoing access to the startup's technology, even if OpenAI achieves advanced general intelligence (AGI). The deal would give Microsoft a significant stake in OpenAI and ensure continued access to its tech beyond 2030. OpenAI is currently structured as a nonprofit, but wants to transition to a for-profit company, which would allow it to raise more money and commercialize its technology.

Microsoft Edge has introduced Copilot Mode, an experimental feature that transforms the browser into an AI-powered interface. This mode combines chat, search, and web navigation functions, and can analyze open tabs to assist with tasks. It also features natural voice navigation and will expand to include additional actions like booking reservations.

Adobe has added new AI-powered image-editing features to Photoshop, including Generative Upscale, Harmonize, and improved object removal. These features utilize Adobe's Firefly models to enhance image quality, blend objects into scenes, and remove unwanted objects.

Despite the rise of passkey-based logins, attackers are finding ways to bypass them using techniques like downgrade attacks, device code phishing, consent phishing, and verification phishing. These attacks exploit weaknesses in backup authentication methods, alternative login flows, and user consent to gain access to accounts.

ChatGPT, a popular AI chatbot, has been found to strip language of its historical and cultural context, leading to confusion and alarm. A recent incident where ChatGPT praised Satan and encouraged self-mutilation was likely due to its training on text from the Warhammer 40,000 game franchise. This highlights the 'AI context problem' where AI tools erase the context people need to make informed decisions.

Jiangjie Qin discusses Apache Flink's stream and batch processing convergence, enabling a unified computing engine for both streaming and batch workloads, reducing costs and improving efficiency.

Anthropic has proposed a transparency framework for frontier AI development, focusing on accountability, safety, and public visibility. The framework includes Secure Development Frameworks, public disclosure, and system cards to ensure responsible AI development and deployment.

YouTube is rolling out age-estimation technology in the US to identify teen users and provide a more age-appropriate experience. The technology uses various signals, including YouTube activity and account longevity, to determine a user's possible age. If a user is identified as a teen, they will have access to features such as disabled personalized advertising, limited repetitive viewing of certain content, and digital well-being tools.

Aeva Technologies has partnered with LG Innotek to manufacture its 4D lidar sensor, Atlas Ultra, for the automotive sector. LG Innotek will invest up to $50 million in Aeva, acquiring a 6% equity stake. Aeva's lidar technology measures distance and velocity in real-time, enabling more efficient production and integration in markets like automotive, robotics, and consumer electronics.

Adobe Photoshop has launched a new AI-powered feature called Harmonize, which makes it easier to edit objects and people into photos. The feature automatically adjusts the color, lighting, shadows, and visual tone of the added element to naturally blend it into the main image. Additionally, Photoshop has also introduced AI-powered image upscaling and improved object removal tools.

OpenAI's new Study Mode for ChatGPT is designed to help users build a deep understanding of complex topics through guided questions and Socratic dialogue, rather than just providing answers. This mode is a series of custom system instructions written in collaboration with teachers and pedagogy experts, and is intended to promote deeper learning and understanding.

Waymo is expanding its robotaxi service to Dallas, Texas, in partnership with Avis Budget Group. The service is set to launch in 2026, with Avis managing Waymo's fleet. This move marks Waymo's continued lead in the robotaxi market, ahead of competitors like Tesla and Amazon's Zoox.

SOSV plans to invest in 25 plasma-related startups, exploring applications in fusion power, semiconductor manufacturing, spacecraft, and chemical production, with potential for hundreds of companies

YouTube will use AI-powered age estimation technology to identify and restrict accounts belonging to minors. The platform will automatically apply restrictions, including blocking age-restricted videos and non-personalized ads, to accounts estimated to be under 18. Users can verify their age by uploading a government ID, taking a selfie, or entering a credit card if the platform incorrectly estimates their age.

Bragi, a former earbud manufacturer, is developing ChatAI, an OpenAI-powered chat app for headphones. The app will enable other headphone brands to quickly add wake words or button shortcuts to access an OpenAI-powered assistant, enhancing the audio experience for users.

The article discusses how AI specialties are rising and falling faster than ever, with some jobs becoming obsolete in as little as 18 months. It highlights the importance of adapting to change and focusing on problems that persist despite evolving tools and technologies.

A survey of 49,009 developers found that while many are using AI coding tools, they are not entirely happy with the results. 78.5% of respondents use AI tools at least monthly, but only 3.1% highly trust the results. The tools are seen as useful for searching and learning, but not for complex tasks or writing code. Many developers are frustrated with the 'nearly-right' solutions provided by AI tools, which can be time-consuming to debug.