NerdNews

OpenAI has partnered with the US National Laboratories to deploy its advanced AI model on a supercomputer, aiming to supercharge scientific breakthroughs in areas like materials science, renewable energy, and national security. The partnership will provide 15,000 scientists with access to OpenAI's latest reasoning models, with a focus on accelerating US global tech leadership and stabilizing the power grid.

Researchers at Stanford and the University of Washington have created a low-cost AI reasoning model rivaling OpenAI's in just 26 minutes, for under $50. The model, called s1, was refined using a small dataset of 1,000 questions and a method known as distillation, which allows smaller models to draw from the answers produced by larger ones. This breakthrough could threaten to upend the entire AI industry, proving that major companies don't need to spend billions of dollars training AI models.

OpenAI has launched Deep Research, a new agent within ChatGPT that conducts in-depth investigations across the web, automating time-consuming research and providing comprehensive insights on complex topics. Powered by a specialized version of OpenAI's o3 model, it is optimized for reasoning, data analysis, and structured research. Initially available to Pro users, with plans to expand access to Plus and Team users.

Researchers at Stanford and the University of Washington have developed a new AI reasoning model called S1, which was trained using less than $50 in cloud compute credits. S1 is a direct competitor to OpenAI's o1 and has shown impressive performance, but its training relied heavily on an existing model, Gemini 2.0. The model's ability to reason and produce answers to prompts is based on a simple yet effective method of 'thinking' through related questions.

Meta plans to release six new mixed reality devices this year, including AI-powered wearables, as part of its 'Year of Greatness' initiative. The company aims to drive sales, retention, and engagement, particularly in mixed reality, and is working on various products, including smart glasses with displays and potentially a new VR headset.

Lyft has partnered with Anthropic to integrate its AI technology into the company's platform. The partnership will see Lyft use Anthropic's Claude chatbot to handle customer complaints, with the goal of improving resolution times and customer experience. The chatbot is already being used to resolve thousands of driver issues daily, with an 87% reduction in average resolution times. The companies plan to expand the use of Anthropic's tech across Lyft's business, with Lyft providing feedback and testing to help improve the AI models.

Wealthsimple built a GenAI gateway to enhance employee productivity while safeguarding data security. They learned lessons about centralizing tools, adopting trends, and being deliberate with GenAI strategy. The company saw increased adoption and engagement with GenAI tools, with over 2200 messages sent daily.

A Lenovo-commissioned report finds that many business leaders are unconvinced about the return on investment for AI, despite significant investments from companies like Microsoft and Google. The report highlights a disconnect between the hype around AI and the actual value it provides to businesses, with 37% of respondents expressing reservations about signing purchase agreements. Only 5% of respondents have adopted AI across their enterprise, and many proof-of-concept projects are stuck in the pilot phase.

According to Context, 40% of laptops sold in Europe in Q4 were AI PCs, but only 5% were Copilot+ PCs. The high price point, 57% higher than average, and software compatibility issues are major factors contributing to the low sales. Microsoft is now targeting enterprise buyers, but app compatibility remains a challenge.

France is hosting the AI Action Summit, shifting the conversation from safety concerns to AI's benefits and opportunities. The summit will focus on five core themes, including AI's use for public good, the future of work, and global governance. Attendees include OpenAI's Sam Altman, Google's Sundar Pichai, and European Commission president Ursula von der Leyen. The summit aims to create a new foundation to ensure AI's benefits are widely distributed and to develop an informal 'Coalition for Sustainable AI' to standardize measures for AI's environmental impact.

Google is introducing SynthID technology to its Magic Editor, which will imperceptibly watermark photos edited using the Reimagine feature. This move aims to increase transparency and label AI-generated work, making it easier to identify edited images. The watermark will be visible by clicking 'About this image'.

Nintendo has announced the Switch 2, a new console that is set to launch in 2025. The console looks similar to the original Switch but with a bigger screen and improved Joy-Con controllers that attach via magnets. Nintendo has also announced a Switch 2 Direct event for April 2nd, where more details about the console and its games will be revealed.

Google is adding digital SynthID watermarks to photos edited using the Magic Editor's generative AI feature. The watermark is intended to help identify images that have been manipulated using AI tools. However, Google admits that some edits may be too small for SynthID to detect.

Amazon EventBridge now allows users to deliver events directly to AWS services in different accounts, simplifying architecture and reducing latency. This feature enables the use of multiple accounts to improve security and business processes, with fine-grained control over event delivery using IAM roles and resource policies.

Researchers found that modifying 30 lines of code in the Linux kernel could cut data center energy consumption by 30-45%. The proposed solution rearranges operations within the Linux networking stack, improving efficiency and performance. This change could reduce energy consumption and lower operating costs for data centers.

The North Korean Lazarus group attempted to target a Bitdefender researcher with a fake job offer on LinkedIn. The scam involved a sophisticated malicious campaign to steal credentials and deliver malware. The attacker sent a fake job offer with vague details and requested the researcher's CV and personal GitHub repository link. The malware downloaded was a cross-platform infostealer that targeted cryptocurrency wallets and collected login data.

A critical remote code execution (RCE) vulnerability in Microsoft Outlook, tracked as CVE-2024-21413, is being exploited in attacks. The bug allows attackers to bypass Outlook's Protected View and execute arbitrary code via malicious Office documents. CISA has added the vulnerability to its Known Exploited Vulnerabilities catalog, warning federal agencies to secure their systems by February 27.

Hackers are targeting SimpleHelp RMM clients, exploiting vulnerabilities (CVE-2024-57726, CVE-2024-57727, and CVE-2024-57728) to create admin accounts, drop backdoors, and potentially lay the groundwork for ransomware attacks. The Sliver post-exploitation framework is being used, allowing attackers to maintain access and execute commands on infected hosts.

Cisco has released patches for two critical vulnerabilities in its Identity Services Engine (ISE) security policy management platform. The vulnerabilities can be exploited by authenticated remote attackers with read-only admin privileges to execute arbitrary commands as root and bypass authorization on unpatched devices. Affected devices should be migrated or upgraded to one of the fixed releases as soon as possible.

Kaspersky researchers have found a malicious app called ComeCome in the Apple App Store and Google Play Store, which contains hidden optical character recognition spyware designed to steal cryptocurrency. The malware, dubbed SparkCat, can extract crypto wallet recovery phrases from screenshots and send them to a remote command-and-control server. The app has been removed from the App Store, but others containing the same malware remain in the Google Play Store.