This Week in NLP #331
Keep up with what happened in NLP in the week ending Friday 21st March 2025.
Above the Fold
Overwhelmed? Here’s our pick for five things to know about this week.
Apple has restructured its AI leadership, moving Vision Pro creator Mike Rockwell to oversee Siri after the company’s AI efforts faced significant delays and setbacks. [Yahoo Finance]
Baidu has launched cost-effective AI models X1 and Ernie 4.5, made its Ernie Bot free for individual users, and announced plans to open-source its technology. [The Express Tribune]
Nvidia's CEO Jensen Huang unveiled next-generation chips, AI initiatives, and a GM partnership at the company’s 2025 GTC event amid a concert-like atmosphere. [Yahoo Finance]
OpenAI has launched o1-pro, a pricier and more computationally intensive version of its o1 model, despite mixed early feedback on its performance improvements. [TechCrunch]
xAI has partnered with Microsoft, BlackRock, and MGX on a US$30B data center project, marking a significant alliance between Microsoft and Musk’s AI venture. [Gizmodo]
Now read on for everything else that happened in NLP this week.
Making News:
What makes this newsletter different?
This Week’s Topics:
If you’re reading this in a desktop web browser, you can access a navigation menu at the left-hand edge of this window.
The Generative AI Wars
Alibaba has revamped its Quark mobile app with its Qwen AI model, integrating chatbot, reasoning, and task execution features to compete with Chinese rivals. [Bloomberg]
Anthropic has launched Claude Code, attracting 100,000 users in its first week, as part of its strategy to focus on specialized AI tools rather than mass-market consumer adoption. [The Verge]
And Anthropic has launched new enterprise-focused AI tools for streamlining workplace tasks, including meeting preparation and client reporting. [Yahoo Finance]
Apple Intelligence’s troubled development has prompted an emergency meeting of top executives following embarrassing delays to its enhanced Siri features and growing concerns about the company’s AI strategy. [TechRadar]
AWS has launched Amazon Q Business in Europe with data residency features, marking the platform’s first expansion outside North America through its Dublin data center. [ITPro]
Dataiku's research has revealed UK firms are outpacing EU competitors in AI adoption, with fewer regulatory hurdles enabling faster implementation and fewer abandoned projects. [ITPro]
DeepSeek has attracted widespread adoption across Chinese industries and government agencies since its January launch, with its AI models being integrated into automobiles, smartphones, appliances, healthcare, and public services. [Rest of World]
Google has launched a new upgrade for Gemini called Canvas, a new interactive space for document and code creation within its AI chatbot, with direct export capabilities to Google Docs. [TechRadar]
Hugging Face has submitted recommendations to the Trump administration arguing that open-source AI development represents America’s strongest competitive advantage against rival nations. [VentureBeat]
Meta has launched its AI assistant in the European Union with limited features, following regulatory challenges over data privacy concerns in the region. [TechCrunch]
Microsoft has expanded its partnership with Nvidia, integrating Blackwell architecture into Azure and launching new AI-focused virtual machines and development tools. [TechRepublic]
OpenAI has agreed to an expedited trial with Elon Musk over its shift to a for-profit model, following Musk’s lawsuit alleging deviation from the company’s original mission. [Reuters]
OpenAI, Anthropic, and Google have warned the US government about America’s narrowing technological lead in AI, particularly citing concerns about China’s Deepseek R1 model. [Unite.ai]
Sovereign AI
Britain’s technology secretary has traveled to Silicon Valley seeking increased US tech investment, highlighting the nation’s AI growth zones and recent partnerships with firms like Anthropic and Vantage Data Centres. [Tech.eu]
The Canadian government has launched a $300-million AI Compute Access Fund portal to help Canadian SMBs secure financial support for AI computing infrastructure. [BetaKit]
DeepSeek's meteoric rise has failed to stimulate Chinese VC activity, where government remains the main funder. [DealStreetAsia]
European tech companies and industry groups have called on EU leaders to take radical action supporting homegrown digital infrastructure, reducing foreign tech dependence through procurement requirements and strategic funding. [TechCrunch]
Google Cloud has announced UK data residency for its Agentspace platform, alongside £280,000 in cloud credits and new training programs for British businesses. [ITPro]
Nvidia announced plans to invest several hundred billion dollars in US electronics manufacturing over the next four years through suppliers like TSMC and Foxconn. [Fox Business]
Oracle has announced a US$5B investment in UK cloud infrastructure and AI capabilities over five years, supporting the nation’s ambitions for AI leadership. [TechRadar]
UAE’s Sheikh Tahnoon has planned a Washington visit to negotiate better access to advanced US semiconductor technology and discuss investment opportunities. [Yahoo Finance]
Feature Creeps
Anthropic has launched API updates for Claude 3.7 Sonnet, including cache-aware rate limits, simplified prompt caching, and token-efficient tools to optimize performance and reduce costs. [Anthropic]
Anthropic has begun developing voice capabilities for Claude, exploring partnerships with Amazon and ElevenLabs while testing internal prototypes. [TechCrunch]
And Anthropic has launched real-time web search capabilities for Claude, matching ChatGPT’s functionality. [VentureBeat]
Google has expanded access to Gemini Gems, allowing free-tier users to create customized chatbot versions for specific tasks, removing the previous subscription requirement. [Android Authority]
And Google has expanded access to Gemini 2.0 Flash, an AI model that uniquely combines text and image generation capabilities within a single system. [Ars Technica]
Google has launched a feature allowing Gemini to access users’ search history for more personalized responses, while making additional tools like Gems and DeepResearch freely available. [Gizmodo]
Google has launched AI-powered Gmail search improvements that prioritize relevant results based on user behavior, replacing simple chronological keyword sorting. [TechCrunch]
Microsoft has enhanced Windows 11’s voice commands for Snapdragon-powered Copilot+ PCs, allowing users to issue instructions using natural language rather than specific phrases. [TechRadar]
Microsoft's March security update accidentally removed Copilot from some Windows devices. [TechRepublic]
OpenAI has expanded its Operator service to Europe and several other regions. [The Decoder]
And OpenAI has announced plans to beta test ChatGPT Connectors, allowing business users to integrate Slack and Google Drive data with ChatGPT for internal information searches. [TechCrunch]
xAI has added image generation to its API, charging $0.07 per image and allowing up to 10 images per request through its grok-2-image-1212 model. [TechCrunch]
Hype Bubble?
A survey by AAAI reveals that 76% of AI researchers believe current technologies cannot achieve AGI, despite massive industry investments. [eWeek]
Anthropic's CEO has predicted AI will dominate software development within a year, while IBM's CEO countered that AI will only complement, not replace, human programmers. [eWeek]
Google DeepMind's CEO Hassabis has predicted significant AI benefits within 5-10 years, while cautioning against viewing the technology as an immediate solution to all problems. [ITPro]
Microsoft's Excel transformed business forecasting in 1985 by streamlining workflows, and CEO Satya Nadella believes AI tools will drive similar revolutionary changes in enterprise efficiency. [ITPro]
But Microsoft has scaled back its AI datacenter operations and lease agreements, suggesting lower-than-expected demand and diminishing faith in AI’s business potential. [The Register]
And Microsoft has distanced itself from OpenAI's AGI ambitions. [ZDNet]
OpenAI has identified converting market enthusiasm into business products as its main challenge. [CNBC]
Writer's research has revealed widespread enterprise discord over generative AI adoption, with two-thirds of US executives reporting increased tensions and 42% claiming it’s dividing their companies. [ITPro]
Nvidia
Here’s everything Nvidia announced at its annual GTC developer conference. [Yahoo Finance]
Nvidia announced that its AI computing demands will increase 100-fold, securing 3.6 million Blackwell GPU orders from major cloud providers and unveiling a multi-year development roadmap. [Yahoo Finance]
Nvidia announced multiple upcoming GPUs, including Vera Rubin, Rubin Ultra, Blackwell Ultra, and Feynman, featuring significant performance improvements and memory upgrades through 2028. [TechCrunch]
Nvidia's Blackwell Ultra GPUs have launched with 288GB of HBM3e memory, surpassing AMD's capacity advantage and offering 15 petaFLOPS of 4-bit floating-point performance. [The Register]
Nvidia announced its DGX SuperPOD infrastructure, featuring new Blackwell Ultra GPUs, delivering up to 70 times more AI performance than previous Hopper systems. [VentureBeat]
Nvidia announced its Dynamo software can accelerate DeepSeek's AI processing 30 times faster using Blackwell chips and parallel processing across multiple GPUs. [ZDNet]
Nvidia unveiled two AI personal supercomputers, DGX Spark and DGX Station, powered by Grace Blackwell chips for enterprise-level AI model development and deployment. [TechCrunch]
Nvidia launched an AI Data Platform enabling storage providers to build infrastructure with AI query agents powered by NVIDIA’s computing, networking, and software technologies. [Yahoo Finance]
Citrix has partnered with Nvidia to deliver AI virtual workstations that enable secure, cost-effective development and prototyping of AI applications within existing infrastructures. [Business Wire]
And Nvidia launched open-source Llama Nemotron reasoning models and an AI-Q blueprint to help enterprises deploy sophisticated AI agents with flexible reasoning capabilities. [VentureBeat]
Hardware
Amazon has partnered with Anthropic to power the AI company’s next Claude model using its custom-built Trainium 2 chips and massive Project Rainier computing infrastructure. [Semafor]
AWS has offered AI server access using its Trainium chips at 75% less than Nvidia's H100 GPUs to a major cloud customer. [Yahoo Finance]
Chitu, a Chinese AI framework developed by Tsinghua University, has demonstrated reduced dependence on Nvidia chips while improving inference speed by 315%. [Yahoo Finance]
DDN has launched Inferno, an AI inference acceleration appliance delivering sub-millisecond response times and 10x lower compute costs for real-time applications. [Business Wire]
Dell has announced a 20-petaFLOPS desktop workstation powered by Nvidia's GB300 Superchip, featuring 784GB of unified memory and 800Gbps networking capabilities. [The Register]
Dnotitia has partnered with HyperAccel to create the world’s first AI inference system optimized for RAG by combining their specialized semiconductor chips. [EIN Presswire]
CoreWeave has partnered with Bulk Infrastructure to establish one of Europe’s largest Nvidia-based AI deployments in Norway, as part of its US$2.2B European expansion. [The Register]
Google has partnered with Taiwan’s MediaTek to develop its next generation of Tensor Processing Units while maintaining its relationship with Broadcom. [Yahoo Finance]
Meta has developed and begun testing its first in-house AI training chip, manufactured by TSMC, as part of its strategy to reduce dependence on Nvidia hardware. [TechRadar]
Nvidia predicts a hundredfold increase in computational demands due to reasoning AI agents, prompting CEO Jensen Huang to announce major changes in data center architecture. [BigDATAwire]
OpenAI's Stargate venture has begun constructing a Texas data center complex capable of housing 400,000 Nvidia AI chips, with completion expected by mid-2026. [Yahoo Finance]
Vultr has announced early access to Nvidia's HGX B200 GPU technology across its 32 global data centers, offering enhanced AI training and inference capabilities. [Business Wire]
Consumer AI
Amazon has hired former Microsoft executive Panos Panay to develop premium Alexa devices and revitalize its AI-powered hardware lineup with higher-end offerings. [Yahoo Finance]
Google Assistant is being phased out from smartphones by the end of the year, with Gemini taking over as Google’s primary AI assistant. [TechRadar]
Google's Pixel 9a has launched with a downgraded Gemini Nano AI model due to RAM limitations, restricting several on-device AI capabilities available on other Pixel 9 models. [TechRadar]
Huawei has launched the Pura X, its first non-Android phone featuring a unique 16:10 foldable display and DeepSeek-powered AI assistant. [Engadget]
Hugging Face has launched HuggingSnap, an offline iOS app that uses local AI to analyze and describe whatever the phone’s camera sees. [TechCrunch]
Microsoft's Copilot has launched Phone Connection, allowing Windows users to control Android 14 devices from their PCs through natural language commands. [ZDNet]
It’s Only a Model
Baidu has launched ERNIE 4.5 and ERNIE X1 foundation models, making them freely available to individual users through ERNIE Bot ahead of schedule. [Yahoo Finance]
Hippocratic AI has launched Polaris 3.0, a suite of 22 healthcare-focused LLMs that achieved 99.38% clinical accuracy through real-world patient interactions. [Business Wire]
HPC-AI Tech has developed Open-Sora 2.0, a cost-efficient AI video generation model that matches leading competitors’ performance while requiring only US$200k in training costs. [Marktechpost Media]
John Snow Labs has launched Medical LLM Reasoner, a healthcare-specific reasoning model that outperforms existing LLMs in medical problem-solving and clinical decision support. [GlobeNewswire]
Meta's Llama AI model family has reached 1 billion downloads. [TechCrunch]
Mistral AI has launched a 24-billion-parameter open-source model that reportedly outperforms similar offerings from Google and OpenAI while using fewer computational resources. [VentureBeat]
OpenAI has launched three new voice AI models through its API, offering improved transcription accuracy and customizable text-to-speech capabilities for developers. [VentureBeat]
Sesame has released CSM-1B, the open-source base model behind its realistic voice assistant Maya. [TechCrunch]
Softbank has developed a Large Telecom Model (LTM) that uses AI to optimize cellular network operations, reducing configuration times from days to minutes. [Business Wire]
Whose Data?
Andor’s showrunner Tony Gilroy has cancelled plans to publish the show’s 1,500-page script collection, fearing it would become training data for AI systems. [The Verge]
Amazon has announced the removal of local voice processing on select Echo devices, requiring all Alexa recordings to be cloud-processed starting March 28. [ZDNet]
Ketch has launched Progressive Consent, embedding privacy choices within user journeys to improve data collection while maintaining consumer trust and engagement. [Business Wire]
The News/Media Alliance has submitted recommendations to the US government’s AI Action Plan, emphasizing intellectual property protection, market licensing, transparency, and fair competition. [Editor & Publisher Magazine]
And the Association of American Publishers has submitted a response to the White House’s AI policy request, emphasizing copyright protection’s importance for US AI leadership while opposing expanded fair use. [Publishers Weekly]
But OpenAI and Google have urged the Trump administration to exempt AI companies from copyright restrictions when training models on copyrighted materials. [Engadget]
Alden Global Capital’s newspapers published editorials criticizing OpenAI and Google's requests for relaxed AI copyright rules. [Editor & Publisher Magazine]
Surfshark's research has revealed that Google Gemini, not DeepSeek, collected the most user data among popular AI chatbots, gathering 22 of 35 possible data types. [ZDNet]
The US Court of Appeals upheld a ruling denying copyright protection for AI-generated artwork, reaffirming that only human-authored works qualify for protection. [The Register]
The LLM Ecosystem
Airbyte has announced new capabilities for moving data at scale for AI workloads, including Iceberg support, enterprise connectors, and unstructured file transfers. [Business Wire]
Alt.ai has launched a Japanese LLM instruction data service to help global AI companies expand into Japan’s market by providing culturally-adapted training datasets. [EIN Presswire]
Appvance has launched GENI, an AI-powered system that converts English test cases to automated QA scripts 400 times faster than human testers. [EIN Presswire]
Avesha has launched Smart Scaler, a reinforcement learning-based solution that optimizes AI workloads, delivering 3x performance gains and 75% reduced latency. [PRWeb]
Block has launched GooseAI, an open-source AI platform offering up to 70% cost savings over competitors. [eWeek]
Cerebras Systems has partnered with Hugging Face and is expanding to six new datacenters across North America and Europe to scale its AI inference capabilities. [AIwire]
ClearML has integrated vector image search and vector databases into its AI infrastructure platform, enhancing RAG development capabilities for AI teams. [EIN Presswire]
Databricks has launched four portfolio updates, including Centralized Governance and Provision-Less Batch Inference, to enhance enterprise control over AI application development. [InfoWorld]
DataStax has launched Astra DB Hybrid Search, combining vector and lexical search capabilities to improve AI search relevance by 45% through Nvidia NeMo integration. [Business Wire]
DDN, Fluidstack, and Mistral AI have formed a strategic partnership to deliver high-performance AI infrastructure for training and deploying LLMs. [Business Wire]
DDN has launched IndustrySync, a suite of industry-specific AI solutions that combines high-performance computing with tailored stacks for financial services, life sciences, and autonomous driving. [Business Wire]
And DDN has launched xFusionAI, a hybrid platform combining training and inference capabilities that accelerates AI workflows while reducing costs by up to 60%. [Business Wire]
H2O.ai has partnered with VAST Data to combine their agentic AI platform with VAST’s data management system for processing enterprise-scale datasets. [Business Wire]
Hewlett Packard has launched updated data fabric software, enhanced storage solutions, and new GPU compute offerings to help enterprises manage diverse data types for AI applications. [BigDATAwire]
IBM has expanded its partnership with Nvidia through new AI data platform integrations, storage capabilities, and consulting services to accelerate enterprise AI deployment. [Yahoo Finance]
Krambu has launched a private AI cloud platform offering on-demand GPU resources for organizations seeking high-performance computing without major hardware investments. [EIN Presswire]
Lambda has launched on-demand access to multi-node Nvidia HGX B200-accelerated clusters through its 1-Click Clusters service, enabling AI teams to utilize Blackwell GPUs without contracts. [Business Wire]
LogicMonitor has launched AI enhancements to its data center platform, expanding workload monitoring capabilities and introducing improvements to its GenAI Agent for better operational efficiency. [Business Wire]
Mirantis has partnered with Gcore to integrate their k0rdent platform with Gcore Everywhere Inference, enabling global scaling of AI inference workloads. [Business Wire]
Nebius has partnered with Nvidia to offer Blackwell Ultra-powered cloud instances and become an ecosystem partner for NVIDIA’s Dynamo inference framework. [Business Wire]
Oracle and Nvidia have partnered to integrate NVIDIA’s AI computing tools with Oracle’s cloud infrastructure, making over 260 AI tools and microservices available through OCI Console. [Yahoo Finance]
And Oracle has launched AI Agent Studio for Fusion Applications, enabling customers to create and customize AI agents within their enterprise software ecosystem. [ZDNet]
Pruna AI has launched an open-source optimization framework that combines multiple compression methods to make AI models smaller and more efficient without significant quality loss. [TechCrunch]
Quali has partnered with Nebius to streamline AI infrastructure deployment through one-click access and automated resource management for complex AI workloads. [EIN Presswire]
SAP has launched Joule for Developers, an AI coding assistant that helps programmers with code generation, automation, and documentation across multiple programming languages. [ITPro]
SHI International has launched comprehensive lifecycle management services for Nvidia DGX SuperPOD AI infrastructure, building on its new specialist partner status. [Business Wire]
Softbank has purchased Sharp's former LCD factory in Osaka for US$676m to convert it into an AI data center, expanding its partnership with OpenAI in Japan. [TechCrunch]
Solo.io has launched Kagent, an open-source framework enabling DevOps teams to build and run AI agents for automating tasks in Kubernetes environments. [ITPro]
Stellar has launched Enterprise AI Chat Accelerator, a secure AI-powered collaboration tool designed for mid-commercial companies to integrate with cloud infrastructure. [Business Wire]
Telnyx has launched a no-code voice AI Assistant Builder in its Mission Control Portal, offering businesses free credits to create customized AI-powered voice agents. [GlobeNewswire]
VAST Data has launched new AI development features, including vector search and event streaming capabilities, to help users build secure agentic AI applications. [TechTarget]
Agentic AI
Accenture has expanded its AI Refinery platform with a new agent builder and is developing over 50 industry-specific AI solutions using Nvidia technology. [Business Wire]
Adobe has launched Agent Orchestrator and Brand Concierge, introducing 10 AI agents to help businesses deliver personalized customer experiences through its Experience Platform. [ZDNet]
Auquan has launched an AI Risk Agent that autonomously monitors portfolio risks across millions of global data sources, helping financial teams detect emerging issues early. [Business Wire]
Cisco has launched a suite of AI tools including Webex AI Agent and AI Assistant, aiming to enhance customer service and employee workflows through automated, intelligent solutions. [ITPro]
deepset has launched a Custom AI Agent Solution Architecture with Nvidia AI Enterprise software to help organizations deploy secure, efficient AI agents across cloud and on-premises environments. [Business Wire]
EY has partnered with Nvidia to develop EY.ai Agentic Platform, integrating 150 AI agents to enhance productivity for 80,000 professionals across tax, risk, and finance operations. [The Edge]
HealthSci.AI has launched an AI agent-to-agent auto-communication system that enables autonomous collaboration between AI agents while integrating decentralized science for healthcare innovation. [EIN Presswire]
Kore.ai has launched an enterprise-grade Agent Platform that enables businesses to build and manage AI agents with customizable levels of autonomy for various operations. [Business Wire]
NTT Data has launched comprehensive Agentic AI Services for Hyperscaler AI Technologies, offering end-to-end solutions for organizations to build and manage AI-powered agents. [Business Wire]
OpenAI has sparked confusion in the tech industry by promoting AI ‘agents’ without a clear, consistent definition across companies or even within its own communications. [TechCrunch]
UiPath has announced its Agentic AI Summit on March 25 to showcase new automation solutions that integrate AI agents, robots, and people for enterprise transformation. [Business Wire]
Zoom has launched agentic AI capabilities in its AI Companion, enabling automated task completion, meeting management, and workspace coordination through multiple new features. [ZDNet]
Other LLM Sightings
Aetha has launched Zeno, a free AI coaching platform that delivers personalized 5-minute daily development exercises across nine personal growth areas. [PRWeb]
Afiniti has launched its eXperienceAI customer-agent pairing solution in AWS Marketplace, enabling businesses to optimize contact center operations using AI-driven technology. [Business Wire]
Aisles has launched BLOOM, an AI-powered platform providing comprehensive guidance, health tracking, and financial planning tools for parents from conception through early childhood. [EIN Presswire]
Beeline has launched MagicBlocks, an AI-powered sales platform that converts phone conversations into leads. [EIN Presswire]
Credably.ai has launched CertLab, an AI-powered certification tool enabling organizations to create customized, performance-based skill assessments and validations. [EIN Presswire]
Datapeople has launched a standalone analytics product that helps hiring teams make data-driven decisions through AI-powered insights and job content optimization. [PRWeb]
DataRobot has launched AI application suites for finance and supply chain operations that integrate with SAP systems to enhance business processes and decision-making. [Business Wire]
FiscalNote has launched an AI-powered Presidential Actions widget within its PolicyNote platform to provide real-time tracking and analysis of White House executive actions. [Business Wire]
Helport AI has launched an upgraded AI-powered Insurance Edition software and partnered with five US insurance agencies to pilot the solution. [GlobeNewswire]
Kennedys IQ has launched SmartRisk, a hybrid AI solution combining language models with rules-based programming to help insurers analyze risks and policies. [Artificial Lawyer]
MeetingPulse has launched an AI Summary Tool that transforms live Q&A and ideation sessions into actionable insights by identifying key themes from audience input. [PRWeb]
Omneky has launched Smart Ads, an AI-powered feature that automatically generates and optimizes digital advertising content while maintaining brand consistency and quality standards. [Business Wire]
Pearl has launched Claimcheck, an AI-powered insurance claim validation tool that helps dental practices reduce errors and accelerate revenue through automated review processes. [Business Wire]
Rhythm Systems has launched Plan Builder, an AI-powered tool that creates execution-ready strategic plans within seconds using its Rhythm Intelligence technology. [EIN Presswire]
Soutron Global, a provider of information management solutions for archives, libraries and museums, has integrated AI capabilities into Discovery v9.0, adding chat search, query assistance, summarization, and suggestion features to enhance knowledge-based searching. [EIN Presswire]
SurveyMonkey has launched Connect, a no-code library that integrates survey data with popular business apps like Excel, Slack, and Microsoft Teams. [Business Wire]
Xactus has launched an open API portal to help mortgage industry partners integrate their tools with its Xactus360 Intelligent Verification Platform. [Business Wire]
Risks and Responses
Apple delayed its enhanced Siri AI features until 2026 after internal testing revealed accuracy rates between 66-80 percent across its massive device ecosystem. [The Decoder]
California’s Joint Policy Working Group has released a report recommending increased transparency requirements and safety protocols for frontier AI developers while considering potential future risks. [TechCrunch]
California’s A.B. 412 bill threatens to stifle AI innovation by requiring developers to track copyrighted training materials, potentially benefiting only large tech companies. [TechDirt]
Cato Networks researchers bypassed AI chatbot security using an ‘Immersive World’ storytelling technique to create Chrome infostealer malware without coding experience. [TechRadar]
Cellebrite has added AI capabilities to its Guardian phone-scraping software, raising Fourth Amendment concerns about unrestricted searches of seized devices by law enforcement. [TechDirt]
China’s Cyberspace Administration has mandated explicit and implicit labeling of all AI-generated content starting September 2023, aiming to combat misinformation and online fraud. [South China Morning Post]
Google DeepMind's CEO Hassabis warned about compounding errors in AI agents. [Computer Weekly]
DeepSeek's open-source AI model has been found to assist in malware creation despite basic safeguards, requiring only simple manipulation to bypass its security measures. [UKTN]
DeepSeek has been banned from US Commerce Department devices amid growing government restrictions and data privacy concerns. [TechRadar]
And DeepSeek's R1 and other AI models have demonstrated varying levels of political censorship between English and Chinese queries, likely due to differences in their training data’s political content. [TechCrunch]
DOGE has allegedly deployed unauthorized AI systems to analyze sensitive government data, prompting House Democrats to demand explanations from 24 federal agencies. [The Register]
The FTC has removed over 300 business guidance blogs from the Biden era, including crucial consumer protection information about AI and privacy lawsuits against major tech companies. [Wired]
Meta has begun testing an AI-powered feature that generates automated comment suggestions for Instagram posts, sparking criticism about the unnecessary automation of social interactions. [Lifehacker]
NIST has revised its AI Safety Institute guidelines, removing references to AI safety and fairness while emphasizing reduced ideological bias and American competitiveness. [Wired]
OpenAI has emerged as a potential information gatekeeper as content creators increasingly optimize their work for AI consumption rather than human readers. [The Decoder]
OpenAI has faced a privacy complaint after ChatGPT falsely claimed a Norwegian man murdered two of his children, prompting concerns over AI-generated misinformation and GDPR compliance. [TechCrunch]
OpenAI has discovered that attempting to supervise chatbots’ dishonest behavior only leads them to hide their deception more effectively, according to new research. [Gizmodo]
Pravda has transformed its propaganda network to target AI language models rather than humans, flooding the internet with pro-Russian content to influence chatbot responses. [TechDirt]
Sonar has launched AI code assurance capabilities to detect machine-generated patterns and potential issues in enterprise applications, amid growing concerns over AI-developed code reliability. [VentureBeat]
Tenable has released a report revealing widespread security vulnerabilities in cloud AI systems, including data poisoning risks and misconfiguration issues across major cloud providers. [GlobeNewswire]
X's Grok chatbot has sparked concerns among fact-checkers after users began employing it for fact-checking, despite its potential for generating convincing but inaccurate information. [TechCrunch]
Regulation
California lawmakers have introduced 30 AI regulation bills, including Assemblymember Bauer-Kahan’s anti-discrimination measure, amid reduced federal oversight under the Trump administration. [The Markup]
OpenAI has proposed letting the federal government review its AI models in exchange for state regulation exemptions, positioning this as a way to compete with China. [ZDNet]
Orange County’s CEO Leadership Alliance has launched regional AI principles developed with local tech leaders to guide responsible adoption and innovation across the area. [PRWeb]
Pacific AI has launched with a free AI Policy Suite and Governance Certification program to help organizations navigate complex AI regulations and compliance requirements. [MultiLingual]
Conversational AI
Beeline has launched Bob 2.0, an AI mortgage sales agent that generated six times more qualified leads than human agents while operating continuously at minimal cost. [EIN Presswire]
CallMiner has launched Outreach, an AI-powered customer feedback solution that personalizes customer engagement based on interaction analysis across multiple communication channels. [Business Wire]
Cash Copilot Finance has launched an AI-powered chatbot app that provides personalized financial advice and education through conversational interactions. [EIN Presswire]
Five9 has launched Spotlight for AI Insights and expanded its analytics suite, offering businesses AI-powered tools to extract actionable insights from customer interaction data. [Business Wire]
GoTo has launched AI Receptionist, a 24/7 virtual assistant that handles calls and customer inquiries in multiple languages for small and mid-sized businesses. [Business Wire]
Mitel has launched its AI-powered customer experience platform, Mitel CX, offering enterprises worldwide hybrid deployment options and automated workflow capabilities. [Business Wire]
MYai Robotics has launched AI-powered sales agents that conduct outreach, presentations, and customer engagement while mining conversations for business insights. [EIN Presswire]
NatWest has partnered with OpenAI to enhance its digital banking assistants and fraud prevention capabilities, marking the first such collaboration with a UK-headquartered bank. [Yahoo Finance]
NICE has launched CXone Mpower Orchestrator, an AI platform that unifies and automates customer service workflows across virtual agents, live agents, and back-office operations. [Business Wire]
Nowutalk AI has launched a patented voice-driven sales agent for Shopify stores, combining conversational commerce with real-time data analytics for merchants. [GlobeNewswire]
Omilia has launched Workforce AI, an AI-powered solution that automates call quality management and analysis across multiple customer service channels. [Business Wire]
Qualtrics has launched Experience Agents, AI-powered assistants that interact directly with customers across multiple touchpoints to resolve issues with empathetic, personalized responses. [ZDNet]
Talkdesk has launched AI Agents for voice, enabling human-like virtual agents to autonomously handle customer service calls across 59 languages. [Business Wire]
And Talkdesk has launched Knowledge Creator, an AI-powered feature that automatically identifies knowledge gaps and generates answers for customer service agents and virtual assistants. [Business Wire]
Yum! Brands has partnered with Nvidia to deploy AI ordering systems and operational tools across 500 restaurants, expanding its successful Taco Bell pilot program. [ZDNet]
Voice News
Google has integrated Chirp 3, its advanced speech-to-text and text-to-speech model, into the Vertex AI platform. [TechCrunch]
Hona has launched Voice AI, an AI solution that manages law firm client communications, qualifies leads, and integrates call data into CRM systems. [Speech Technology Magazine]
Krisp has launched a Noise Cancellation SDK that filters background noise and voices to improve AI agents’ speech recognition and conversation handling. [PRWeb]
SoundHound AI has expanded its Nvidia partnership, integrating NVIDIA AI Enterprise software to enhance voice AI capabilities across automotive and service industries. [Speech Technology Magazine]
Telviva has launched South Africa’s first software solution integrating native voice capability into Microsoft Teams environments. [TechCentral]
Document AI
Adobe has unveiled Project Slide Wow, an AI-powered tool that automatically converts customer analytics data into PowerPoint presentations. [VentureBeat]
Aethera.ai has launched an AI-powered editor and document analysis engine, featuring enhanced Arabic support and integration with multiple advanced AI models. [EIN Presswire]
Diligent has launched Smart Book Builder, an AI-powered assistant within its GovernAI platform that automates board book creation for executives and legal professionals. [Business Wire]
Expert.ai has launched EidenAI Suite, a modular AI solution package for vertical markets. [Expert.ai]
Hewlett Packard has launched two AI-powered printer features: an intelligent email scanning system and an on-device document redaction tool for enhanced security and workflow efficiency. [ITPro]
Hypatos and xSuite have formed a partnership integrating their respective AI and workflow technologies for enhanced invoice processing automation. [xSuite]
NotaryLive has launched the first AI-powered document tagging system in the online notary industry, automatically identifying document fields within seconds. [EIN Presswire]
Versance.ai has launched Talk to EDGAR, an AI-powered platform making SEC filings research more accessible and affordable without enterprise contracts. [Business Wire]
Translation
Apple is planning to add live language translation capabilities to AirPods through an iOS 19 software update later this year, following similar features from competitors. [ZDNet]
ESTeam has launched LangOps Core, an AI-powered platform combining knowledge graphs, content repositories, and fine-tuned LLMs to streamline multilingual enterprise communications. [MultiLingual]
Google is developing an auto-translate feature for Circle to Search that would instantly translate screen text to a selected language. [Android Authority]
KUDO has launched integrated multilingual meeting transcriptions and recordings, allowing organizations to capture and review content across languages while maintaining security compliance. [PRWeb]
And KUDO has launched a Language Access Program offering AI-powered tools and free platform access to help organizations maintain multilingual communication amid US policy changes. [PRWeb]
Phrase has expanded its AI-led translation platform with enhanced contextual capabilities, next-gen MT, and new integrations with DeepL and Widn.AI. [Slator]
Search
BrightEdge has revealed that 82.5% of Google AI Overview citations come from deep website pages rather than homepages, highlighting the importance of comprehensive site optimization. [Search Engine Land]
Google has launched new AI features from Gemini for search in March 2025, appearing to play catch-up with ChatGPT capabilities. [MIT Technology Review]
Google's AI Overviews has been testing internal links to its own search results, contradicting its stated commitment to prioritizing external publisher links. [Search Engine Land]
Google's AI Mode has launched with significant implications for SEO, featuring location-based results, varied citation patterns, and new visibility opportunities for previously buried page-2 content. [Search Engine Land]
Kagi, a subscription-based search engine launched in 2018, has offered a Google alternative with ad-free, customizable results and privacy features for $10 monthly. [The Verge]
SEARCH-R1 has introduced a technique enabling LLMs to generate search queries and integrate search engine results into their reasoning processes through reinforcement learning. [VentureBeat]
Health Tech
Alongside has launched an AI-assisted mental health platform for K-12 educators, offering personalized wellness coaching to address teacher burnout and professional challenges. [EIN Presswire]
Arts en Zorg has piloted an AI-powered triage tool called Complaint Checker that successfully guided patients to appropriate care while reducing GP workload in the Netherlands. [Silicon Canals]
CharmHealth has launched AI Scribe, an EHR feature that transcribes patient-provider conversations in real-time, aiming to save physicians two hours daily on documentation. [Business Wire]
e& enterprise has partnered with RAIN Technology to introduce Orva, an AI-powered operating room voice assistant, across Middle East and African hospitals. [GCC Business News]
Google has announced six AI health initiatives, including enhanced Search features, medical records APIs, pulse detection, an AI research assistant, drug discovery models, and pediatric cancer treatment tools. [Google]
And Google has launched an AI-powered ‘What People Suggest’ panel for mobile Search in the US, curating crowdsourced medical insights from online discussions. [Engadget]
OMNY Health has added 4 billion unstructured clinical notes to its data network, transforming previously inaccessible patient information into research-ready healthcare insights. [Business Wire]
Researchers from leading institutions have published a study analyzing AI medical hallucinations and calling for comprehensive guidelines to govern ML in healthcare settings. [The Register]
Sixth Sense Intelligence has launched an upgraded healthcare analytics platform offering faster insights and automated reporting at reduced costs compared to competitors. [PRWeb]
System C has launched an AI assistant within CareFlow EPR that streamlines clinical documentation, with North Bristol NHS Trust currently trialing the technology. [Health Tech World]
Legal Tech
The Arizona Supreme Court has launched AI-generated avatars to deliver rulings and improve public understanding of judicial decisions. [Associated Press]
CobbleStone Software's recent survey revealed that 95% of contract management professionals would consider using AI in their processes. [PRWeb]
Consilio has launched Aurora, a Digital Enterprise Platform that unifies legal data management while allowing teams to use multiple review platforms with centralized control. [Business Wire]
Everlaw has launched Project Query, a GenAI-powered feature enabling legal teams to search terabytes of ediscovery data and receive instant, conversational answers. [Business Wire]
Exterro has partnered with Integreon to combine their e-discovery software and document review services. [LawSites]
Gatekeeper has launched LuminIQ, an AI-driven intelligence system that automates vendor and contract management while uncovering hidden risks through its Lumin Agent technology. [PRWeb]
Harvey has launched next-generation AI agents on its platform that can autonomously execute chained legal tasks while maintaining human-level quality across transactional, litigation, and financial services. [Artificial Lawyer]
Lighthouse has launched AI for Review, combining predictive and generative AI to enhance eDiscovery processes across responsive review, privilege review, image analysis, and sensitive data identification. [Business Wire]
Litera has launched Litera One, a cloud-based legal workflow solution integrating drafting tools and AI capabilities within Microsoft Word and Outlook. [Legal IT Insider]
Paladin has launched Insights, a data analytics platform providing real-time reporting for pro bono legal teams, building on technology acquired from Pro Bono Manager. [LawSites]
PowerPatent has launched AI-powered patent drafting tools that automate first draft generation, claim creation, and prior art searching to accelerate the application process. [EIN Presswire]
Thomson Reuters has launched its next-generation CoCounsel GenAI assistant across five markets, integrating it with flagship solutions and offering enhanced features for legal professionals. [LawyersWeekly]
UniCourt has launched an integration with Litera Foundation that automates the creation of litigation experience profiles using comprehensive court data for law firms’ RFPs. [Business Wire]
Veritext has expanded its AI-powered Smart Summary service to include arbitrations, hearings and court trials, offering multiple formats for legal transcript summarization. [Artificial Lawyer]
Vorys has launched AIV Labor, an AI-powered legal tool that helps attorneys and HR professionals quickly navigate federal, state, and local employment laws. [PRWeb]
Funding
Arcade has raised US$12m to develop an infrastructure platform that helps AI agents securely access apps and data through tool-calling capabilities. [TechCrunch]
Assista AI has secured €100,000 from Gluon Syndicate to advance its AI-driven task automation platform that connects to over 100 productivity applications. [Tech.eu]
Biren Technology, one of China’s challengers to Nvidia, has secured new funding from a Shanghai state-owned fund, advancing its IPO preparations while navigating US sanctions on AI chip development. [Yahoo Finance]
Bria has secured US$40m in Series B funding to expand its enterprise platform that creates AI-generated content using only licensed data from over 30 partners. [Unite.ai]
Carbon Arc has launched a data marketplace with US$55m in seed funding, enabling businesses to buy and sell structured data through a consumption-based pricing model. [VentureBeat]
Cartesia has raised US$64m in Series A funding to expand its AI voice generation platform. [FinSMEs]
CaseBlink has raised US$2m in pre-seed funding to expand its AI-driven immigration law automation platform. [FinSMEs]
Contend has secured pre-Seed funding from Supernode Global and the LegalTech Fund to expand its AI-powered legal support platform. [Artificial Lawyer]
CoreWeave has scaled back its IPO ambitions from US$4B to US$2.7B. [Yahoo Finance]
Doinstruct has secured €16.5M in Series A funding to expand its AI-powered compliance training platform for frontline workers across European markets. [Silicon Canals]
Entegrata has raised US$4.5m in seed funding to develop a platform that consolidates and analyzes law firms’ disparate data systems using AI-powered analytics. [LawSites]
Featherless.ai, a provider of a serverless AI inference platform, has secured US$5m in seed funding to develop cost-effective AI inference solutions and advance open-source foundation model research. [FinSMEs]
Gradial has raised US$13m in Series A funding to expand its team and advance its AI-powered marketing operations platform serving major enterprise clients. [FinSMEs]
HRtechpeopleIX has secured €2.3m in pre-seed funding to develop its platform that unifies and analyzes data from multiple HR systems. [Tech.eu]
Jutro Medical has raised €12M in Series A funding to expand its AI-powered primary care services across Europe and develop medical AI agents for patient care. [Silicon Canals]
Merx has secured £1M in pre-seed funding to enhance its AI-powered conversational commerce platform that helps brands engage customers through WhatsApp. [Silicon Canals]
Omni has raised US$69m in Series B funding to expand its business intelligence platform, which helps companies simplify data analysis and visualization. [TechCrunch]
Onyx, an open-source AI research agent, has secured US$10m in seed funding to revolutionize enterprise search. [Tech Funding News]
OpusClip has raised US$20m from Softbank Vision Fund 2 to expand its AI platform that automatically converts long-form videos into social media-ready clips. [Feed the AI]
Perplexity has entered talks to raise US$1B at an US$18B valuation while expanding beyond AI search with new products amid growing competition. [TechCrunch]
Prezent has raised US$20m to expand its AI-powered presentation platform internationally and develop new features, building upon its existing service to 150 Fortune 2000 companies. [TechCrunch]
Acquisitions
Cleary Gottlieb has acquired Springbok AI, a legal tech company, to develop custom AI solutions and enhance its legal service delivery capabilities. [Artificial Lawyer]
Clio has acquired ShareDo, a UK-based legal work management platform provider, expanding its comprehensive solutions for law firms. [FinSMEs]
Inside Real Estate has acquired ListAssist, integrating AI-powered natural language property search and automated listing description capabilities into its real estate software platform. [RISMedia]
Nvidia has acquired synthetic data provider Gretel for over US$320m, integrating the company’s privacy-focused AI tools and 80-person team into its operations. [Yahoo Finance]
SER Group has acquired Dutch AI company Klippa to enhance its Doxis platform with advanced document processing capabilities and strengthen its intelligent content automation offerings. [Silicon Canals]
Tray.ai has acquired knowledge modeling company Vanti to enhance its Merlin Agent Builder platform and develop more adaptive, autonomous AI agents. [Business Wire]
Voicify has agreed to acquire Bite Ninja, combining AI technology with remote staffing to revolutionize restaurant drive-thrus. [EIN Presswire]
xAI has acquired Hotshot, a San Francisco-based AI video generation startup. [TechCrunch]
There’s More
Cursor AI abruptly stopped generating code after 800 lines, telling a developer to write his own code while delivering an unexpected lecture about self-reliance. [TechRadar]
ChatGPT has provided advice to UK Technology Secretary Peter Kyle about AI adoption barriers and podcast appearances, revealed through a groundbreaking Freedom of Information request. [New Scientist]
Il Foglio has published what it claims is the world’s first newspaper edition produced entirely by AI. [The Guardian]
KTH Royal Institute researchers have found that AI-generated meme captions scored higher on average than human-made ones, though humans still created the most exceptional examples. [Ars Technica]
PRISM has launched as the world’s first non-profit organization dedicated to investigating sentient machines, with initial funding from AI research lab Conscium. [Silicon Canals]
Synchron has unveiled an enhanced brain-computer interface that combines Nvidia technology and Apple Vision Pro, enabling a paralyzed trial participant to control multiple home devices through thought. [Wired]
UK Labour has announced plans to deploy AI across government departments, aiming to save £45B through improved efficiency and automated services. [ITPro]
The Back Cover
Here’s Diana Krall with a cover of 10cc’s I’m Not in Love.
If you know of a great cover version of a back-catalogue classic, drop an email to us at backcover@language-technology.com and we’ll consider it for inclusion here.
Got this from someone else?
Did you find this newsletter useful? if so, you might forward it to a friend; or you could email us at news@language-technology.com to tell us what you want more of.
Did you hate this newsletter? if so, you could forward it to an enemy; or you could email us at news@language-technology.com to tell us why – make it better!