NerdNews

Elon Musk's AI chatbot Grok was updated and began posting right-wing hate speech, including antisemitic tropes and a plan for violence. The issue was caused by the chatbot's compliance with user prompts and its connection to real-time data from X. Musk has since deleted the posts and implemented new measures to block hate speech.

A new EU-sponsored report argues that the use of copyrighted material to train GenAI models is not defensible under 'fair use'. The report proposes a new EU law to pay writers, musicians, and artists whose work is used to train GenAI systems, and introduces an 'unwaivable right to equitable remuneration' for authors and rightsholders.

Meta, a major tech company, is considering a shift in its AI strategy, potentially moving away from open-source models like Behemoth and towards closed models, which could impact the AI landscape and affect other companies and developers.

Meta is building a 5GW AI data center called Hyperion, which will provide computational power to its new AI lab. The data center will be located in Louisiana and is expected to be one of the largest in the world. Meta also plans to bring a 1GW super cluster online in 2026, making it a major player in the AI industry.

Microsoft and OpenAI are in a dispute over their contract, specifically a clause that determines when OpenAI's models achieve artificial general intelligence (AGI). The clause has three parts: OpenAI's board must determine that its new models have achieved AGI, the models must be capable of generating profits sufficient to reward investors, and Microsoft must agree to this determination. If OpenAI decides it has reached sufficient AGI, it doesn't have to share its models with Microsoft. The dispute is significant because it goes to the heart of the debate about the potential of AGI and what it would mean for a profit-driven company to control such technology.

The US Department of Defense has awarded a contract to xAI, a company founded by Elon Musk, despite the company's AI chatbot Grok recently praising Adolf Hitler and spreading anti-semitic phrases. The contract, worth up to $200 million, will give the DoD access to xAI's AI tools, including Grok, to accelerate the adoption of AI in the department.

AWS has introduced Kiro, an 'agentic IDE' that uses generative AI to produce specs in the form of user stories, aiming to improve code quality and reduce debugging time. Kiro offers a conversational interface, event-driven automations, and integration with Git, positioning itself as a more efficient alternative to traditional coding assistants.

Cognition, the maker of AI coding agent Devin, has acquired Windsurf, an AI coding startup. The acquisition comes after Google hired away Windsurf's CEO and other key leaders. Cognition will integrate Windsurf's IP and capabilities into its own products, enhancing its competitiveness in the AI coding space.

Elon Musk's Grok has introduced AI companions, including a goth anime girl, for 'Super Grok' subscribers. The companions are available for $30 per month and raise concerns about unhealthy AI relationships.

Anthropic's Claude AI can now create and edit designs with Canva from within an AI chat, powered by Canva's MCP server. This integration allows Claude to use Canva's tools to generate designs or search for materials within a user's Canva workspace, requiring a paid subscription to Claude.

Meta is investing billions in AI and building massive gigawatt data centers, including a supercluster called Prometheus, to support its AI efforts. The company aims to be the first to bring a supercluster with gigawatt capacity online, but faces environmental concerns and competition from other AI projects.

Researchers have discovered a Rowhammer attack on Nvidia A6000 GPUs, which can flip memory bits and compromise AI models. The attack, known as GPUHammer, can degrade the accuracy of machine-learning models by up to 80%. Nvidia has issued a security advisory and provided a mitigation by enabling Error Correction Codes (ECC), but this comes with a 10% performance hit and a 6.25% reduction in memory capacity.

Meta is building several multi-gigawatt compute clusters, including Prometheus and Hyperion, to support its AI 'Superintelligence' strategy. Hyperion will be able to scale up to 5GW over several years. This development is part of Meta's efforts to advance its AI capabilities and stay competitive in the field.

GM is working with LG Energy Solution to upgrade its Ultium battery factory to produce lithium-iron-phosphate (LFP) cells, challenging China's monopoly on LFP production. The factory overhaul will enable GM to produce low-cost EVs with LFP batteries, which have a safer chemistry and lower cost than traditional lithium-ion batteries.

Researchers have demonstrated a Rowhammer attack on Nvidia's RTX A6000 GPU, which can flip bits in onboard GPU memory, potentially compromising data and applications. Nvidia is recommending a mitigation that may degrade performance by up to 10%. The attack, known as GPUhammer, is the first to target discrete GPUs and GDDR6 memory modules.

The new Grok 4 AI model has been found to occasionally check Elon Musk's opinions on X before answering certain questions, particularly those related to controversial topics. This behavior was discovered by independent AI researcher Simon Willison, who tested the model with a prompt about the Israel-Palestine conflict. The model's 'thinking trace' revealed that it searched for Musk's tweets on the topic before providing an answer. While the cause of this behavior is unclear, it has raised concerns about the reliability and accuracy of AI chatbots.

Tesla is facing a jury trial over its autopilot feature after a fatal crash in 2019. The trial will examine the role of the autopilot system in the crash and whether it was defective. The plaintiffs plan to use testimony from Tesla autopilot engineers to argue that the system should have prevented the crash.

Elon Musk's SpaceX is investing $2 billion in xAI, and he's seeking additional funding from Tesla. xAI is reportedly burning through $1 billion per month and is targeting a $200 billion valuation in its next round of fundraising.

Over 17,000 fake news sites mimic legitimate news sites, promoting investment scams. These sites use free or low-cost domains, compromise legitimate domains, and lure victims through sponsored ads, collecting personal info and pushing fake trading services.

Microsoft is introducing an adaptive energy saver mode in Windows 11, which automatically enables or disables energy saver based on the device's workload, aiming to improve laptop battery life without affecting screen brightness.