NerdNewsFebruary 13, 2025 |
News & Trends
Anima Anandkumar Highlights AI’s Potential to Solve ‘Hard Scientific Challenges’
Anima Anandkumar, a professor at California Institute of Technology, is using AI to help solve the world’s challenges faster. She has used neural operators, a type of AI framework, to speed up prediction models for extreme weather and work on sustainable nuclear fusion simulations. Her work has led to the development of FourCastNet, an AI-based high-resolution weather model that is tens of thousands of times faster than traditional weather models.
OpenAI Lays Out Plans for GPT-4.5 and GPT-5
OpenAI CEO Sam Altman has shared a roadmap for the company's upcoming AI models, GPT-4.5 and GPT-5. GPT-4.5, also known as Orion, will be the last non-chain-of-thought model, and GPT-5 will integrate various technologies, including o3. The company plans to simplify its product lineup and provide unlimited chat access to free ChatGPT users with the release of GPT-5.
Adobe launches AI video generator
Adobe has launched Firefly, an AI video generator that can create video from text prompts and still images. The model provides control over shot size, camera angle, and camera motion, making it useful for professionals. It can generate up to 5 seconds of footage at 1080p resolution and 24fps. Firefly is available in a new web app and offers two subscription plans: Standard and Pro.
Google to use machine learning to identify underage users
Google will test a machine learning-based age estimation model to predict if a user is under 18 and apply age filter settings. The feature will use data like search history and video categories to make predictions. Google will also roll out School Time, a feature to limit screen time, to Android phones and tablets.
Larry Ellison wants to put all America's data in one big Oracle system for AI to study
Larry Ellison suggests that the US should put all its data, including DNA and healthcare records, into one big Oracle system to be analyzed by AI. He believes this will lead to better healthcare and more efficient governance, but raises concerns about surveillance and privacy. |
Options & Tutorials
ChatGPT may not be as power-hungry as once assumed
A new study by Epoch AI suggests that ChatGPT's energy consumption is approximately 0.3 watt-hours per query, lower than the previously estimated 3 watt-hours. This is due to the use of more efficient AI models like GPT-4o. However, as AI usage expands, energy demands are expected to increase, with AI data centers potentially requiring enormous power capacity in the coming years.
Efficient DevSecOps Workflows With a Little Help From AI
Michael Friedrich discusses how AI can be used to improve DevSecOps workflows, including code creation, testing, and security scanning. He highlights the importance of establishing guardrails for AI and measuring its impact on workflows.
Feds want devs to stop coding 'unforgivable' buffer overflow vulnerabilities
The FBI and CISA have labeled buffer overflow vulnerabilities as 'unforgivable defects' and are urging developers to adopt secure-by-design practices to avoid creating them. They recommend using memory-safe coding languages such as Rust, Go, and Swift, and implementing a phased transition plan to rewrite entire codebases. The agencies also suggest leveraging technologies to limit memory safety vulnerabilities and conducting aggressive adversarial product testing.
CHERI Security Hardware Program Essential to UK Security
The UK government is backing the CHERI security hardware program to address rising cyber risks. The program, based on the Capability Hardware Enhanced RISC Instructions (CHERI) project, aims to secure underlying computer hardware and prevent memory safety software vulnerabilities. With £80m in government funding and £200m from industry, the program has developed effective secure hardware devices and tools. The technology has the potential to significantly reduce cybersecurity costs and improve customer experience.
Leveraging Open-source LLMs for Production
Andrey Cheptsov discusses the practical use of open-source large language models (LLMs) for real-world applications, highlighting their pros and cons, and providing insights into their development process, optimization techniques, and use cases. |
Launches & Tools
Google backs Boston quantum startup QuEra in $230M debt round
QuEra, a Boston-based quantum computing startup, has raised $230 million in a convertible note led by Google. The company plans to use the funding to build a 'useful' fully quantum computer in the next three to five years. QuEra is working on a neutral atom quantum supercomputer, which relies on using lasers to cool atoms in the computing process to reduce errors. The company has already generated revenue through the sale of a quantum computer to Japan and offers cloud services on AWS.
Australian health tech startup Harrison.ai scores $112M Series C
Harrison.ai, an Australian health tech startup, has raised $112 million in Series C funding to scale its international reach. The company offers AI-enabled medical diagnostic software and services to analyze CT scans, X-rays, and pathology slides for radiologists and medical examiners.
Russia's Sandworm caught stealing credentials
Russia's Sandworm hacking group, linked to the Russian Military Intelligence Unit 74455 (GRU), has been caught stealing credentials and data from organizations in the US, UK, Canada, and Australia. The group, also known as Seashell Blizzard, has been carrying out a 'near-global' initial access campaign dubbed 'BadPilot' since at least 2021, exploiting vulnerabilities in software such as Microsoft Exchange, Zimbra Collaboration, and OpenFire.
Google State Cybercrime Report
Google's Threat Intelligence Group reports that cybercrime and state-sponsored hacking are increasingly linked, with states relying on criminal capabilities to further their missions. The report calls for lawmakers to take action, including designating cybersecurity as a national security priority and incentivizing the implementation of best practices. Google also recommends enhancing international cooperation and disrupting the cybercrime ecosystem.
AI summaries turn real news into nonsense, BBC finds
The BBC has published research showing that AI assistants, including OpenAI's ChatGPT, Microsoft's Copilot, Google's Gemini, and Perplexity, often provide inaccurate news summaries. The research found that 51% of AI answers to news questions had significant issues, with 19% introducing factual errors and 13% altering quotes from BBC articles. The findings highlight the challenges of using generative AI for news summarization. |
Quick Links
Oracle Makes Fusion Apps Available on EU Sovereign Cloud
Oracle has made its Fusion Cloud Applications Suite available on its EU Sovereign Cloud, offering users greater assurance in compliance with the region's data law. The move is designed to address the needs of public sector organizations and regulated industries that require data sovereignty. The Fusion Cloud Applications Suite includes ERP, HCM, Supply Chain & Manufacturing, and Customer Experience applications.
Google’s I/O developer conference set for May 20-21
Google's annual developer conference, I/O, will take place on May 20-21, 2025, at the Shoreline Amphitheater in Mountain View. The event will feature a keynote by CEO Sundar Pichai and various breakout sessions for developers, with a focus on AI and Google's generative AI platform, Gemini.
Apple's robotics won't hit mass production until 2028
Apple's robotics won't be available for mass production until 2028. The company is working on robots for the smart home, both in non-humanoid and humanoid forms, but they are still in the internal proof-of-concept stage. Apple is focusing on anthropomorphic creations like the lamp concept instead of human-like hardware, prioritizing sensing hardware and software as core technologies.
US, UK and Australia Sanction Russian Bulletproof Hoster Zservers
The US, UK, and Australia have imposed joint sanctions on Zservers, a Russian bulletproof hosting provider, and six individuals for supporting ransomware groups like LockBit. Zservers allegedly provided hosting services for command-and-control servers, data leak sites, and other malicious activities.
IBM Return-to-Office Mandate
IBM has implemented a return-to-office mandate for its Finance & Operations business unit, requiring employees to live within 50 miles of designated hubs or face severance. The move is seen as a 'soft layoff' and is part of the company's efforts to reduce costs and invest in AI, with plans to replace thousands of jobs with automation. |
Share NerdNewsShare your affiliate link to get commission!
Thanks for reading,
If you dont want to receive future editions of NerdNews, |