Last Updated on February 19, 2025 by Editor
As we step into Week #7 of 2025, the AI landscape continues to evolve at a rapid pace. From global policy discussions to innovative AI tools and major investments, the past week has been filled with significant developments shaping the future of artificial intelligence. This roundup covers key highlights such as South Korea’s AI infrastructure boost, strategic AI investments, and the launch of cutting-edge tools like DeepHermes-3 and AI-powered mobile apps. As the AI sector advances, these updates offer valuable insights into the technological and regulatory trends that will define the industry’s trajectory in the coming years.
Elon Musk’s xAI Unveils Grok-3: The Smartest AI on Earth
On February 17, 2025, Elon Musk’s artificial intelligence company, xAI, officially launched Grok-3, its latest AI chatbot designed to compete directly with industry leaders like OpenAI’s ChatGPT and emerging Chinese AI startup DeepSeek. Musk has described Grok-3 as the “smartest AI on Earth,” highlighting its advanced reasoning capabilities, superior computational power, and future integration with Tesla vehicles.
Grok-3 represents a significant leap in computational resources, having been trained using 10 times the compute of its predecessor, Grok-2. This training took place at a state-of-the-art data center in Memphis, utilizing 200,000 GPUs to enhance its learning and reasoning capabilities. In performance benchmarks, Grok-3 achieved a 52% score on the AIME 2024 Benchmark, surpassing OpenAI’s GPT-4o, Claude 3.5 Sonnet, and Gemini 2.0 Pro. It also attained a 75% score on the GPQA Science Benchmark, showcasing its advanced scientific reasoning and problem-solving skills. Additionally, Grok-3 introduces innovative features such as a self-correcting mechanism, DeepSearch technology for transparent thought processes, and plans for integration with Tesla vehicles to provide real-time AI interaction via voice commands. Currently, Grok-3 is available exclusively to Premium+ subscribers on X, with future updates planned to enhance its capabilities further.
South Korea Blocks Downloads of DeepSeek Amid Privacy Concerns
South Korea’s data protection authority has suspended new downloads of the Chinese AI app DeepSeek due to the app’s failure to fully comply with the country’s personal data protection regulations. The suspension, effective since February 15, 2025, will remain until DeepSeek addresses the necessary privacy law standards. While new downloads are halted, existing users can still access the service. The Personal Information Protection Commission (PIPC) initiated this action following DeepSeek’s acknowledgment of partial neglect in adhering to the regulations. This move aligns with similar actions in other countries; for instance, Italy’s data protection authority ordered the blocking of DeepSeek’s chatbot over privacy issues in January 2025. The Chinese government has emphasized its commitment to data privacy and legal compliance, though DeepSeek has not yet commented on the suspension.
OpenAI Revises Model Specifications to Address Controversial Topics
OpenAI has released a significant update to its Model Specification (Model Spec), expanding it to 63 pages and making it freely available for public use and modification. The updated guidelines emphasize customizability, transparency, and intellectual freedom, aiming to improve how AI models handle controversial topics. OpenAI encourages its models to engage in truthful discussions while maintaining moral stances on issues like misinformation and potential harm. The revised spec provides a framework for responsibly managing sensitive queries, integrating feedback from ongoing AI ethics debates. Notable changes include nuanced approaches to mature content and addressing “AI sycophancy.” Released under a Creative Commons Zero license, OpenAI invites public feedback to refine the behavior of its AI models further. This update coincides with OpenAI’s announcement of the upcoming GPT-4.5 model, reflecting the company’s efforts to align its models’ behavior with evolving societal standards and ethical considerations.
Apple Intelligence Set to Enhance Vision Pro in 2025
Apple is planning to enhance its Vision Pro headset by incorporating Apple Intelligence, introducing a guest user mode, and adding a spatial content app as early as April 2025. These updates will be part of a software upgrade, potentially available in beta form for developers soon. Apple is focusing on integrating artificial intelligence (AI) into its devices to stay competitive, engaging in talks with Tencent and ByteDance to integrate their AI models into iPhones sold in China. The Vision Pro, priced at $3,499, is expected to feature new AI-driven tools such as Writing Tools, Genmojis, and the Image Playground app. This initiative aims to revitalize sales of the device, which have declined due to its high cost and competition from more affordable alternatives like Meta Platforms’ Quest.
Perplexity AI Introduces Freemium ‘Deep Research’ Tool for Comprehensive Analysis
Perplexity AI has unveiled its latest feature, ‘Deep Research’, designed to provide users with in-depth research and analysis capabilities. This tool aims to deliver expert-level insights across various domains, including finance, marketing, and technology, in under three minutes. Users can access ‘Deep Research’ for free with limited daily queries, while Pro subscribers enjoy unlimited access. Currently available on the web, Perplexity plans to expand this feature to iOS, Android, and Mac platforms soon.
In performance evaluations, ‘Deep Research’ achieved a 21.1% score on the ‘Humanity’s Last Exam’ benchmark, surpassing models like Gemini Thinking and Grok-2, though it trails behind OpenAI’s 26.6% score. This launch positions Perplexity as a competitive player in the AI-driven research sector, offering a cost-effective alternative to premium services.
Amazon and Apple’s AI Voice Assistants Face Developmental Delays
Both Amazon and Apple are encountering challenges in enhancing their AI voice assistants, Alexa and Siri, with generative AI capabilities. Amazon’s anticipated AI-powered Alexa upgrade, initially slated for a February release, has been postponed to at least March 31 due to issues with inaccurate responses during testing. Similarly, Apple’s overhaul of Siri is experiencing engineering problems and software bugs, potentially delaying its rollout.
These setbacks highlight the complexities involved in integrating advanced AI functionalities into existing voice assistant frameworks. The delays may impact the competitive landscape, providing an advantage to rivals like OpenAI’s Advanced Voice Mode and Google’s upcoming Gemini Live, which are advancing in the AI voice assistant market.
Elon Musk’s xAI Seeks $10 Billion Funding at $75 Billion Valuation
Elon Musk’s artificial intelligence startup, xAI, is reportedly in discussions to raise $10 billion in a funding round that would value the company at approximately $75 billion. Existing investors, including Sequoia Capital, Andreessen Horowitz, and Valor Equity Partners, are expected to participate. This follows previous funding rounds totaling $12 billion in 2024.
The additional capital is intended to expand xAI’s data center capacity for further AI model training. In a related development, xAI is nearing a deal to purchase over $5 billion in AI servers from Dell Technologies, equipped with Nvidia’s GB200 chips. These strategic moves underscore xAI’s commitment to scaling its AI capabilities and competing with industry leaders.
EU Denies Dropping AI Liability Rules Under US Pressure
The European Union has refuted claims that it abandoned its proposed AI Liability Directive due to pressure from the Trump administration. The directive, introduced in 2022, aimed to establish common rules for compensating individuals harmed by AI-enabled products and services. EU officials assert that the decision to withdraw the proposal was driven by internal considerations to reduce bureaucratic hurdles and enhance competitiveness, not external influence.
Despite shelving the directive, the EU remains committed to responsible AI governance. Plans are underway to introduce a code of practice linked to the broader AI Act, focusing on streamlining reporting requirements and supporting AI innovation within the region. This approach reflects the EU’s intent to balance technological advancement with consumer protection in the evolving AI landscape.
UK Rebrands AI Safety Institute and Partners with Anthropic
The UK government has rebranded its AI Safety Institute to the AI Security Institute, shifting its focus toward addressing serious AI-related security threats such as cyberattacks, chemical and biological weapons development, fraud, and child sexual abuse. This strategic move, announced by Technology Secretary Peter Kyle at the Munich Security Conference, emphasizes the UK’s commitment to safeguarding national security and protecting citizens from malicious AI applications. The institute will collaborate closely with the Home Office, establishing a “criminal misuse” team to tackle these pressing issues.
In conjunction with this rebranding, the UK has signed a memorandum of understanding with AI company Anthropic. This partnership aims to explore the integration of Anthropic’s AI assistant, Claude, into public services to enhance accessibility and efficiency. The collaboration will focus on developing best practices for the responsible deployment of advanced AI technologies within the public sector, ensuring that AI advancements align with societal values and contribute positively to public service delivery.
Anthropic’s Next Major AI Model Set for Imminent Release
AI research firm Anthropic is poised to unveil its next significant AI model in the coming weeks. This new model is expected to combine traditional language processing capabilities with advanced reasoning features, excelling particularly in coding tasks and offering developers enhanced control over performance parameters. The hybrid approach allows the model to function as both a standard language model and a deep reasoning engine, adapting to diverse use cases on demand. Notably, the model has demonstrated superior performance on certain programming tasks compared to OpenAI’s o3-mini-high “reasoning” model.
The forthcoming release underscores Anthropic’s commitment to advancing AI technology while maintaining a focus on safety and ethical considerations. By integrating enhanced reasoning capabilities, the new model aims to provide more accurate and contextually relevant outputs, thereby broadening the scope of AI applications across various industries. This development positions Anthropic as a formidable competitor in the rapidly evolving AI landscape.
YouTube Shorts Enhances Content Creation with Veo 2 Integration
YouTube has integrated Google DeepMind’s Veo 2 video generation model into its Dream Screen feature for YouTube Shorts, enabling creators to generate AI-powered video clips using simple text prompts. This advancement allows users to create standalone video content or enrich existing Shorts with dynamic, AI-generated visuals, thereby expanding creative possibilities without the need for additional footage or complex editing tools.
The Veo 2 model enhances the realism and detail of generated videos by better understanding real-world physics and human movement. Creators can customize outputs by specifying styles, lenses, or cinematic effects, making content creation more accessible and versatile. Currently, these features are available to users in the U.S., Canada, Australia, and New Zealand, with plans for broader expansion in the future. All AI-generated content is labeled with SynthID watermarks to ensure transparency regarding AI involvement in content creation.
EnCharge AI Secures Over $100 Million to Revolutionize AI Processing
EnCharge AI, a semiconductor startup specializing in analog in-memory computing chips for AI applications, has raised over $100 million in a Series B funding round led by Tiger Global. The company’s innovative approach integrates analog processing within memory chips, significantly reducing energy consumption for AI inference tasks. EnCharge AI’s technology enables accelerators to perform AI computations with up to 20 times less energy than leading digital AI chips, making it particularly advantageous for edge computing devices like laptops and smartphones.
The infusion of capital will facilitate the commercialization of EnCharge AI’s first products, expected to launch later this year. By addressing the growing energy demands of AI processing, EnCharge AI aims to make AI applications more sustainable and accessible across various platforms, from data centers to consumer electronics. This development highlights a significant shift toward energy-efficient AI solutions in the semiconductor industry.
Latent Labs Secures $50 Million to Advance AI-Driven Programmable Biology
Latent Labs, an AI-driven biotechnology startup founded by Dr. Simon Kohl, a former DeepMind researcher, has emerged from stealth with $50 million in funding. The company aims to revolutionize drug discovery by developing AI foundation models that make biology programmable, enabling the computational creation of new therapeutic molecules such as antibodies and enzymes. This approach seeks to expedite drug development and enhance precision in targeting diseases. The funding round was co-led by Radical Ventures and Sofinnova Partners, with participation from notable investors including Google’s Chief Scientist Jeff Dean and Cohere co-founder Aidan Gomez. Latent Labs operates from offices in London and San Francisco, bringing together a team of experts from leading tech and biotech firms. The company plans to collaborate with pharmaceutical partners to leverage its generative AI platform, aiming to reduce reliance on traditional experimental methods and accelerate the development of effective therapeutics.
Meta Platforms in Talks to Acquire South Korean AI Chip Startup FuriosaAI
Meta Platforms, the parent company of Facebook and Instagram, is reportedly in discussions to acquire FuriosaAI, a South Korean AI chip startup. Founded in 2017 by former Samsung and AMD engineer June Paik, FuriosaAI specializes in developing AI accelerators designed to optimize AI applications. The company’s flagship product, the RNGD AI inference chip, offers a thermal design power of 150W and claims a threefold improvement in performance per watt compared to Nvidia’s H100 GPUs. This potential acquisition aligns with Meta’s strategy to enhance its AI hardware capabilities and reduce dependence on external suppliers like Nvidia. The deal could be finalized as early as this month, reflecting Meta’s commitment to advancing its in-house AI infrastructure.
Drata Acquires SafeBase for $250 Million to Enhance Security Compliance Automation
Drata, a San Diego-based security compliance automation firm, has acquired SafeBase, a trust center platform provider, in a deal valued at $250 million. This strategic acquisition aims to integrate SafeBase’s AI-powered security review automation into Drata’s existing platform, offering a comprehensive solution for governance, risk, and compliance. The combined capabilities will enable organizations to enhance transparency, streamline security assessments, and simplify audit processes. Drata, launched in 2020, has rapidly expanded its customer base to over 7,000 organizations, including notable clients like Notion and Tenable. The acquisition of SafeBase is part of Drata’s aggressive growth strategy, marking its third acquisition in the past year, and underscores its commitment to providing robust security and compliance solutions in an increasingly complex regulatory landscape.
Keychain Raises $5 Million to Expand AI-Driven Manufacturing Platform in Europe
Keychain, an AI-powered manufacturing platform co-founded by Irish entrepreneur Oisin Hanrahan, has secured $5 million in funding to support its expansion into the European market. The investment round was led by Bright Pixel Capital, the venture arm of European retail conglomerate Sonae. Keychain’s platform connects over 30,000 manufacturers with more than 20,000 brands and retailers, streamlining the traditionally fragmented process of sourcing packaging materials and ingredients. By leveraging AI, Keychain aims to enhance transparency and efficiency within the consumer packaged goods (CPG) manufacturing ecosystem. The new funding will facilitate the launch of two new platforms focused on packaging and ingredients, addressing the growing demand for sustainable and efficient production methods. Keychain plans to officially enter the European market later this year, further solidifying its position as a transformative force in the manufacturing industry.
Vice President JD Vance Advocates for Light AI Regulation at Paris Summit
At a recent summit in Paris, U.S. Vice President JD Vance cautioned global leaders against imposing heavy regulations on artificial intelligence (AI), suggesting that over-regulation could stifle innovation and hinder technological progress. Vance emphasized the importance of a balanced approach, advocating for policies that encourage AI development while addressing potential risks. He highlighted that collaborative efforts between nations are essential to harness AI’s benefits effectively.
Thomson Reuters Prevails in Landmark AI Copyright Infringement Case
In a significant legal decision, Thomson Reuters has won a major AI copyright infringement lawsuit against startup Ross Intelligence. The case centered on allegations that Ross Intelligence unlawfully reproduced content from Thomson Reuters’ legal research platform, Westlaw. U.S. Circuit Court Judge Stephanos Bibas ruled in favor of Thomson Reuters, stating that Ross’s defenses were insufficient and that the copying did not constitute fair use. This ruling sets a precedent for how AI companies utilize copyrighted material in training their models.
Elon Musk’s DOGE Develops GSAi Chatbot to Modernize U.S. Government Services
The Department of Government Efficiency (DOGE), led by Elon Musk, is developing GSAi, a custom generative AI chatbot for the U.S. General Services Administration (GSA). This initiative aims to enhance the productivity of the GSA’s 12,000 employees by streamlining tasks such as contract management and procurement analysis. The project aligns with President Donald Trump’s AI-first agenda to modernize federal operations through advanced technology.
Adobe Launches Firefly AI Video Generator Offering IP-Safe Content Creation
Adobe has introduced its Firefly Video Model, an AI-powered video generation tool designed to provide intellectual property-safe content for professional creators. Entering public beta, the model is trained exclusively on licensed content, ensuring that generated videos are free from IP infringement concerns. Integrated into Adobe’s Premiere Pro software, Firefly allows users to create 1080p video clips from text prompts or images, featuring controls for camera angles and atmospheric effects. Subscription plans start at $9.99 per month, making AI-driven video creation more accessible to a broad range of users.
Nous Research Unveils DeepHermes-3: A Toggle-On Reasoning AI Model
Nous Research has introduced DeepHermes-3, an advanced AI model that seamlessly integrates traditional language processing with enhanced reasoning capabilities. This innovative model allows users to toggle between concise responses and detailed, chain-of-thought reasoning outputs, catering to diverse application needs. DeepHermes-3 is built upon Meta’s Llama 3.1 architecture and has been fine-tuned to excel in complex role-playing, strategic planning, and decision-making tasks. Notably, it supports multi-turn conversations and maintains long-term context coherence, making it adept at handling intricate dialogues. The model is accessible for testing on platforms like Hugging Face, with optimized versions available for consumer-grade hardware, ensuring efficient performance across various devices.
SAP Collaborates with Databricks to Launch Business Data Cloud
In a strategic move to enhance AI readiness, SAP has partnered with Databricks to introduce the Business Data Cloud (BDC). This fully managed SaaS solution aims to unify and govern all SAP and third-party data, providing a harmonized foundation for advanced analytics and AI applications. By integrating Databricks’ data engineering and machine learning capabilities, BDC enables organizations to seamlessly connect their SAP data with external datasets, facilitating comprehensive insights and fostering innovation. Early adopters, such as multinational chemical company Henkel, have leveraged BDC to unlock the value of their data, driving business innovation and ensuring long-term adaptability.
PIN AI Launches Mobile App for Personalized, Private AI Models
PIN AI, a startup founded by AI and blockchain experts from Columbia, MIT, and Stanford, has launched a mobile application that empowers users to create personalized, private AI models directly on their smartphones. This app allows individuals to select and fine-tune open-source AI models, such as DeepSeek or Llama, ensuring that all personal data remains on-device and under the user’s control. By eliminating reliance on cloud-based processing, PIN AI addresses privacy concerns and offers a decentralized alternative to traditional AI assistants. The platform also incorporates blockchain technology for secure authentication, providing users with a transparent and customizable AI experience.
As AI continues to transform industries worldwide, the updates from February 12–18, 2025 showcase the incredible strides being made in both technology and policy. With major innovations in AI models and global collaborations, it’s clear that the momentum will only build. Stay tuned for more updates, as the world of AI is ever-evolving. Don’t miss out—check back next week for more insights and key developments that are shaping the future of artificial intelligence!
