NerdNews

Anthropic is throttling usage of its Claude model during peak hours, while OpenAI has removed caps for its Codex model, potentially changing the future of model pricing and usage for AI technologies.

The NeurIPS AI research conference has become embroiled in a geopolitical clash between the US and China. The conference's organizers announced and then quickly reversed new restrictions for international participants, which would have affected researchers at Chinese companies like Tencent and Huawei. The controversy reflects the increasingly fraught political landscape that top researchers must navigate, and could strain ties between US and Chinese researchers.

Google's new TurboQuant algorithm can make Large Language Models' memory usage six times smaller, reducing energy consumption and potentially allowing powerful AI models to run on smartphones. This breakthrough could disrupt the AI economy and impact companies like NVIDIA, which relies heavily on data center infrastructure.

OpenAI has introduced plugin support to its Codex coding app, allowing for easier integration with external services and applications. This move aims to close the gap with competitors like Claude Code and Gemini. The plugins are bundles that include skills, app integrations, and Model Context Protocol servers, making it possible to configure Codex for specific tasks and replicate them across multiple users.

The AI hype is real, and we're not prepared for what's next. AI systems can now do things that used to require human programmers, and they're getting better every year. However, there are huge risks involved, including the potential for large-scale cyber attacks, misinformation campaigns, and other forms of disruption. The alignment problem, which refers to the challenge of making sure AI systems do what we want them to do, is a major concern. Experts are warning that we're entering dangerous territory with AI, and it's essential to slow down and understand what we're building before it's too late.

OpenAI's Sora AI video generator, which allowed users to create realistic videos using text prompts, has been shut down due to high computing costs and lack of profitability. The shutdown affects a $1bn deal with Disney and the API available to developers. This move may indicate a shift in focus towards more serious AI endeavors, but the future of AI video generation remains uncertain.

JumpCloud's platform manages identity, access, and devices to tackle shadow AI. IT teams need a unified view of all activity to monitor and control AI effectively. A strong identity and access management foundation is vital for visibility and security.

The article provides a step-by-step guide to building a custom Bloomberg terminal using AI, starting from simple earnings previews with ChatGPT to more sophisticated systems using Claude with skills and data connectors.

Google has warned that advances in quantum computing could render traditional encryption methods obsolete by 2029, posing a significant threat to cybersecurity. This could allow attackers to break current cryptographic standards, putting sensitive information at risk. To counter this, Google is implementing post-quantum cryptography (PQC) in its upcoming Android 17 operating system.

WebTransport is a new protocol that aims to improve WebSocket performance by reducing latency and enabling transparent network switching. It leverages HTTP/3 and the QUIC protocol to support reliable byte streams and unreliable datagrams, making it suitable for use cases such as high-frequency financial data streaming, cloud gaming, and live streaming.

Bluesky's new app, Attie, is an AI assistant that lets users build custom feeds using natural language. It's powered by Anthropic's Claude and built on top of Bluesky's AT Protocol. Attie will eventually allow users to vibe code their own apps on top of the protocol.

TikTok's advertising policies require AI-generated content to be labeled, but some companies like Samsung are not disclosing AI usage in their ads, raising concerns about transparency and honesty in advertising

A class action lawsuit has been filed against Google by victims of Jeffrey Epstein, alleging that the company's AI Mode feature published personal information about them, including contact details and addresses. The lawsuit claims that Google failed to remove the information despite being notified, and that the AI Mode feature is not a neutral search index but an active recommender and content generator.

The US and UK are seeking technology to counter the threat of underwater drones, with a focus on sensors, counteraction technologies, communications, and command and control systems. The tender, called REEF, has an April 3 deadline and is open to companies of all sizes. The goal is to identify potential defense solutions to protect ships, harbors, and other critical maritime infrastructure from underwater drones.

Waymo's robotaxis are growing in number, but they can get stuck and rely on taxpayer-funded public services to move them. This raises questions about the company's reliance on first responders and the potential risks and challenges associated with autonomous vehicles.

ChatPlayground AI is a Chrome extension that allows users to access multiple AI models in one spot, streamlining their AI workflow. It offers features such as side-by-side results, prompt engineering, and unlimited monthly messaging. A lifetime subscription to the Unlimited Plan is available for $67.15, a discount from the regular price of $619.

The Oppo Find N6 features a nearly creaseless display, clever multitasking software, and impressive battery life, making it the best foldable phone on the market. It has a 6,000mAh battery, 16GB of RAM, and 512GB storage, with a seven-core Qualcomm Snapdragon 8 Elite Gen 5 processor. The phone also supports a stylus and has a triple rear camera with a 200-megapixel main lens.

The UK government has sanctioned a Chinese crypto marketplace, Xinbi, for its alleged role in funding scam centers in Southeast Asia. Xinbi is accused of selling scam victim data and facilitating the laundering of stolen crypto assets. The UK is the first country to sanction Xinbi, which has inflows of over $19.7 billion.

Dutch football club AFC Ajax has admitted to a data breach after an attacker gained access to its internal systems, exposing email addresses and limited personal data. The breach allowed hackers to transfer season tickets, alter account details, and even lift stadium bans, potentially affecting over 300,000 registered supporters and 42,000 season tickets.

Security researchers analyzed 10 million websites and found almost 2,000 API credentials publicly exposed, including those for AWS, GitHub, and Stripe, which could grant access to critical infrastructure and services.