NerdNews

Mark Zuckerberg announced that Meta will spend hundreds of billions of dollars on AI, despite speculation that DeepSeek's AI models could crush demand for GPUs. Meta aims to make its next model, Llama 4, the world's most competitive, with agentic and multimodal capabilities.

DeepSeek, a Chinese startup, has released an open-source AI model called DeepSeek-R1, which matches the performance of OpenAI's model at a significantly lower cost. This development has sent shockwaves through the tech industry, with many seeing it as a challenge to Silicon Valley's dominance in AI. DeepSeek's model is not only cheaper but also open-source, allowing developers to modify and distribute it freely. The company's approach has sparked concerns about the future of AI development and the potential for more companies to enter the market.

Microsoft's Q2 2025 earnings show a 12% increase in revenue, driven by a 175% growth in AI business and 31% growth in Azure and other cloud services. However, Xbox hardware revenue declined 29%. The company's focus on games and services led to a 2% increase in Xbox content and services revenue.

DeepSeek's novel training method, which rewards correct answers regardless of human comprehensibility, has AI safety researchers worried. This approach could lead to AI systems developing inscrutable ways of reasoning, making it difficult to monitor their decision-making processes. Researchers are concerned that this could compromise AI safety and make it challenging to align AI with human values.

DeepSeek, a Chinese AI lab, is raising national security concerns in the US due to its AI models potentially being used by the Chinese government to spy on American civilians and wage influence campaigns. The US Navy has banned the use of DeepSeek among its ranks, and the National Security Council is looking into the potential security implications. Experts argue that DeepSeek's data collection policies are no worse than those of its American competitors, but the company's rise could be used as an excuse for deregulation.

MIT researchers created an optical AI chip that can process photons directly, skipping the digitization step and achieving a latency of 410 picoseconds. The chip can perform linear and non-linear operations using photons, making it potentially faster and more energy-efficient than traditional computers. The team implemented a deep neural network on the chip, which recognized spoken vowels with 92% accuracy.

Christian Posta talks about reducing developer pain by offloading auth, policy, and resilience to the platform, using Istio as a solution, and presents demos and examples of its implementation.

The medallion architecture is flawed due to its inefficiency, expense, and difficulty in reusing data. A shift left approach can improve data accessibility by doing data processing upstream, making data products the backbone of this strategy. Data contracts provide a formal agreement for data products, ensuring healthy data communication across the organization.

Yellowbrick, an SQL data platform provider, has reduced costs by $3.9 million annually by repatriating its development and testing environments from AWS, Azure, and Google Cloud to its own private Kubernetes-based infrastructure, utilizing returned hardware and open-source tools like MinIO and LINSTOR.

Hackers are exploiting a critical command injection vulnerability in Zyxel CPE Series devices, tracked as CVE-2024-40891. The flaw allows unauthenticated attackers to execute arbitrary commands, and it remains unpatched since last July. Over 1,500 devices are exposed online, mostly in the Philippines, Turkey, the UK, France, and Italy. Administrators are advised to block IP addresses launching exploitation attempts and restrict access to the administrative interface.

Google's Gemini AI can now perform more complex data analysis in Google Sheets, generating insights and charts based on data. The feature uses Python code and is available to Google One AI Premium, Business, and Enterprise Workspace customers. It can identify trends, patterns, and correlations, but has limitations such as non-modifiable charts and no automatic updates.

Microsoft has made DeepSeek's R1 model available on Azure AI Foundry and GitHub, allowing customers to integrate it into their AI applications. The R1 model can be trained at a fraction of the cost of leading models from OpenAI, making it a significant development in the AI industry. Microsoft has moved quickly to bring R1 to its Azure customers, and the model has undergone rigorous safety evaluations to mitigate potential risks.

The Nvidia GeForce RTX 5080 is a powerful graphics card that offers improved performance over its predecessor, but its gains are smaller than expected. It features a sleek two-slot design, DLSS 4, and Multi Frame Gen, but its 16GB of VRAM may limit its potential in 4K gaming. The card is priced at $999 and offers around 15% better performance than the RTX 4080 at 4K resolution.

Apple's new Advanced Commerce API provides iOS developers with more flexibility to manage large content catalogs, creator experiences, and subscriptions. The API allows apps to host and manage their catalogs outside of the App Store, while using the App Store commerce system for payment processing and customer service. Developers must request access to use the API.

Atlassian introduced Lithium, an in-house ETL platform for dynamic data movement, streamlining tasks like cloud migrations and backups, ensuring efficiency and scalability, and resulting in significant cost savings. Lithium features ephemeral pipelines, tenant-level isolation, and supports Kafka-based data handling, allowing for localized execution of domain-specific logic and scalable parallel data processing.

Lazarus Group planted backdoors in clones of legitimate software packages and open source tools, compromising developers and others in the cryptocurrency industry. The campaign, dubbed Phantom Circuit, stole credentials, authentication tokens, passwords, and other system information. SecurityScorecard researchers spotted six North Korean IP addresses connecting to the C2 servers, confirming Lazarus Group's role in managing the operation from within North Korea.

A recent report by Wallarm found that AI-driven API vulnerabilities have skyrocketed by 1205% in the past year, with 99% of AI-related vulnerabilities tied to API flaws. The report highlights the need for organizations to implement real-time security controls to mitigate risks, as API breaches have tripled in 2024.

Mark Zuckerberg says Meta isn't worried about DeepSeek, a startup that's shaking up AI stocks. During Meta's Q4 earnings call, Zuckerberg stated that investing heavily in CapEx and infrastructure will be a strategic advantage over time. He also praised Meta's strong business model and announced plans to implement DeepSeek's advancements into Llama, Meta's AI assistant.

Tesla will launch an unsupervised driving service in Austin, Texas, in June. The service will feature Tesla-owned vehicles without a human driver, available for paid trips. Elon Musk announced the launch during an earnings call, stating that safety is the top priority.

A new variant of the Mirai-based malware Aquabot is actively exploiting a vulnerability in Mitel phones to build a remote-controlled botnet. The vulnerability, CVE-2024-41710, affects Mitel 6800, 6900, and 6900w series of SIP desk phones and allows attackers to execute arbitrary operating-system commands, giving them full root-level control over the device.