This Week in NLP #348
Keep up with what happened in NLP in the week ending Friday 18th July 2025.
Above the Fold
Overwhelmed? Here’s our pick for five things to know about this week.
Meta plans to invest hundreds of billions of dollars in AI infrastructure and talent. [Yahoo Finance]
Nvidia has become the world’s first US$4T company, driven by surging demand for its AI chips. [TechRepublic]
And Nvidia has received Trump administration approval to sell its H20 AI chips to China, following months of export restrictions and intensive lobbying efforts. [Yahoo Finance]
OpenAI has launched ChatGPT Agent, an AI tool that combines its Operator and Deep Research capabilities to perform complex, multi-step tasks using a virtual computer. [The Verge]
Thinking Machines Lab has secured a record-breaking US$2B seed investment from major tech investors, following its establishment by former OpenAI executives and researchers. [Wired]
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
Amazon has delayed launching its web-based Alexa.com assistant beyond its planned June release date. [The Washington Post]
AWS is launching an AI agent marketplace, with Anthropic as a partner, enabling startups to sell AI agents directly to AWS customers. [TechCrunch]
Amazon has doubled its investment in AWS Generative AI Innovation Center to US$200m, helping businesses implement AI solutions. [GeekWire]
And AWS has launched Kiro IDE, a conversational development tool that generates detailed specifications before code, aiming to improve upon the limitations of vibe coding. [The Register]
Anthropic has implemented stricter usage limits on Claude Code without warning, frustrating users of its $200 Max plan and causing widespread confusion about service accessibility. [TechCrunch]
The Department of Defense has awarded US$200m contracts to Anthropic, Google, OpenAI, and xAI for AI services, despite concerns over xAI’s Grok chatbot’s controversial behavior. [Gizmodo]
Google has acquired Windsurf's co-founders and technology through a US$2.4B ‘hackquisition’ deal after Microsoft blocked OpenAI's attempted US$3B purchase of the coding startup. [Spyglass]
Google has gained significant market share in cloud computing, challenging Amazon's long-held dominance in the industry. [The Information]
Meta's Superintelligence Lab has considered abandoning its open-source Behemoth AI model in favor of closed development, marking a potential strategic shift from Zuckerberg’s previous stance. [TechCrunch]
Meanwhile, Meta has partnered with AWS to provide $200,000 in cloud credits and technical support to 30 US startups building AI tools using Meta’s open-source Llama model. [CNN]
Microsoft has partnered with San Francisco to provide AI tools to 30,000 city workers, following successful trials showing five-hour weekly productivity gains. [CNBC]
But Microsoft's Copilot has attracted only 79 million downloads compared to ChatGPT’s 900 million, despite being pre-installed on Windows and integrated with Microsoft 365. [Gizmodo]
Nvidia has reached a historic US$4T market value while facing increased Senate scrutiny over Chinese business dealings and competition from rival chipmakers. [Inventiva]
OpenAI has developed a payment checkout system for ChatGPT that will take commissions from online product sales, partnering with Shopify to present the system to brands. [Reuters]
OpenAI has expanded its cloud infrastructure partnerships beyond Microsoft to include Google Cloud, CoreWeave, and Oracle, addressing increased computing demands for ChatGPT and other services. [Yahoo Finance]
OpenAI's advisory board has recommended maintaining nonprofit control over the company, citing AI’s societal impact as too significant for corporate governance alone. [Associated Press]
Perplexity's CEO has warned startup founders that successful AI innovations will be copied by tech giants, citing his company’s experience with web-search features. [TechSpot]
SpaceX has invested US$2B in Elon Musk’s xAI as part of a US$5B equity raise, marking its largest investment in another company. [The Wall Street Journal]
Tesla has launched Grok as a hands-free conversational AI assistant in eligible vehicles, marking xAI's first direct integration with Tesla products. [Teslarati]
xAI is seeking additional funding in a deal potentially valuing the company at up to US$200B, with Saudi Arabia’s PIF expected to be a major investor. [Yahoo Finance]
AI Supremacy
China has launched an aggressive open-source AI strategy, backed by massive state investment and infrastructure, challenging US dominance while raising concerns about authoritarian influence. [Inventiva]
Malaysia has implemented new export controls requiring 30-day advance notice for shipping US-origin AI chips, amid concerns about smuggling to China. [TechCrunch]
Manus AI, the Chinese AI firm known for its advanced AI agent, has relocated its headquarters to Singapore while maintaining it isn’t seeking to circumvent US chip restrictions. [South China Morning Post]
Nvidia CEO Jensen Huang has received a bipartisan warning from US senators about meeting with certain Chinese companies during his planned China trip. [Yahoo Finance]
Nvidia has reversed its planned withdrawal from China and is seeking to restart H20 AI chip sales, linked to rare-earth element trade discussions. [TechCrunch]
Nvidia's CEO dismissed concerns about China’s military using American AI technology, stating they would avoid US tech like America avoids Chinese tech. [Tom’s Hardware]
Nvidia's multi-billion-dollar AI chip deal with the UAE has been suspended due to US concerns about potential chip smuggling to China. [Yahoo Finance]
And Huang has endorsed America’s technology re-industrialization efforts while discussing AI’s future impact on manufacturing, jobs, ethics, and healthcare. [CNN]
Sovereign AI
DDN has partnered with Polarise to establish sustainable, sovereign AI data centers across Europe, combining DDN’s AI platform with Polarise’s Nvidia-powered infrastructure solutions. [Business Wire]
Forrester has outlined Asia’s critical 12-18 month window to establish AI governance frameworks, emphasizing ‘graduated sovereignty’ and strategic adaptation to the AI-driven ‘Seventh Wave’. [FutureIoT]
Google Cloud has launched UK-based data processing for Gemini 2.5 Flash AI, though support remains global and US authorities may still access data. [The Register]
Korean tech companies have launched domestically-developed LLMs focused on Korean language processing, aiming to establish sovereign AI capabilities against global competitors. [BusinessKorea]
Latam-GPT, a Latin American-focused AI language model developed by over 30 regional institutions, has been created to address ChatGPT’s limitations in understanding local languages and cultures. [Rest of World]
Nvidia has partnered with the University of Bristol to launch the UK’s Isambard-AI supercomputer. [Yahoo Finance]
Talkdesk has launched its UK Regional Cloud deployment, enabling British customers to host their platform locally while meeting data privacy requirements and improving voice quality. [GlobeNewswire]
The UK’s Department for Science, Innovation and Technology has launched a £1m Regional Tech Booster programme to support tech entrepreneurs and clusters outside London through mentoring, funding and skills development. [Computer Weekly]
WIRED describes how America’s allies are seeking alternatives to US-controlled platforms after Trump’s administration began weaponizing military, financial, and technological infrastructure against them. [Wired]
Feature Creeps
Anthropic has launched a new feature enabling Claude to directly interact with third-party apps like Canva, Asana, and Figma through connectors for automated data handling. [ZDNet]
Anthropic has launched an analytics dashboard for its Claude Code AI programming assistant, enabling engineering managers to track team usage metrics and ROI of their AI investments. [VentureBeat]
And Anthropic has launched Claude for Financial Services, a specialized version of its enterprise AI platform featuring increased rate limits, pre-built data connectors, and a prompt library. [VentureBeat]
Google has launched Gemini API’s Batch Mode, offering asynchronous processing for large-scale tasks at half the cost of synchronous APIs, with results delivered within 24 hours. [Google]
Google has expanded its NotebookLM platform with featured notebooks from The Economist, The Atlantic, and other publishers, including Shakespeare’s complete works and expert guides. [ZDNet]
Google has rolled out the Gemini 2.5 Pro model and Deep Search features to AI Mode subscribers, while introducing AI-powered business calling for all US users. [Google]
And Google DeepMind has launched GenAI Processors, an open-source Python library that streamlines the development of complex AI applications using LLMs. [Google]
Microsoft has begun testing a Windows 11 feature that enables Copilot AI to view and discuss users’ entire desktop content when explicitly granted permission. [Windows Central]
Mistral AI has enhanced its Le Chat chatbot with deep research capabilities, multilingual reasoning, project organization, and advanced image editing, bringing it closer to leading AI competitors. [TechCrunch]
OpenAI has expanded its Record mode feature to all ChatGPT Plus subscribers on Mac, enabling automated meeting recording, transcription, and summarization capabilities. [ZDNet]
Slack has launched extensive AI features for enterprise productivity, including writing assistance and contextual search, challenging Microsoft's workplace AI dominance. [VentureBeat]
Hype Bubble?
A study by METR revealed that AI coding tools slowed developers by 19% despite their belief that the tools had improved their speed by 20%. [The Register]
CSIRO’s Data61 has examined AI’s complex relationship with productivity, finding mixed results across individual, organizational, and national levels despite widespread optimism about its potential benefits. [Startup Daily]
F5's research has revealed that only 2% of enterprises are highly AI-ready, with most organizations facing security and governance challenges in scaling AI operations. [Business Wire]
Gartner has projected global IT spending to reach US$5.43T by 2025, driven largely by generative AI investments despite economic uncertainties and mixed project outcomes. [ITPro]
MIT researchers have published a paper examining the challenges of AI in software engineering, highlighting gaps between current capabilities and the goal of meaningful automation. [MIT News]
OutSystems has released a study showing 93% of software executives are developing or planning to develop custom AI agents within their organizations. [Business Wire]
Uber founder Travis Kalanick claimed on the All-In podcast that xAI's Grok chatbot was approaching scientific breakthroughs. [Gizmodo]
And concerns about AI’s slowing progress and high costs threaten the chip industry’s future. [Yahoo Finance]
Data Centres
CoreWeave has announced a US$6B commitment to build an AI-focused data center in Lancaster, Pennsylvania, creating hundreds of jobs and strengthening US technological competitiveness. [Business Wire]
Google has announced a US$25B investment in data centers and AI infrastructure, plus US$3B for hydropower plant modernization. [CNBC]
Habitat AI has secured US$16m in seed funding to develop renewable energy-powered infrastructure for autonomous AI systems across Europe, starting with a Netherlands facility. [EIN Presswire]
Hailanyun has launched a $223-million underwater data center project off Shanghai’s coast, using seawater cooling and wind power to reduce energy and freshwater consumption. [Scientific American]
KRAMBU has partnered with Supermicro through an MOU to develop sustainable AI data centers using liquid cooling and renewable energy in Montana and Pennsylvania. [EIN Presswire]
Meta has announced plans to build Manhattan-sized datacenter clusters, with its first 1GW+ facility, Prometheus, scheduled to begin operations in 2026. [TechRadar]
xAI has entered negotiations with Saudi Arabia’s Humain and another firm to lease data center capacity, seeking cheaper energy and political advantages. [Bloomberg]
The Chips are Up
AMD and Broadcom have positioned themselves to potentially outperform Nvidia in AI markets through inference processing and custom chip design respectively. [The Motley Fool]
AMD has partnered with XMPro to deliver an integrated edge AI solution that processes industrial data 8-9 times faster than CPU-only systems while maintaining complete data sovereignty. [EIN Presswire]
CoreWeave has launched Nvidia RTX PRO 6000 Blackwell Server Edition instances on its cloud platform, offering up to 5.6x faster LLM inference performance. [Yahoo Finance]
Huawei has approached Middle Eastern and Southeast Asian buyers about purchasing its Ascend 910B AI chips, challenging Nvidia's market dominance in those regions. [DealStreetAsia]
And Huawei has developed a new AI chip design mirroring Nvidia's architecture to help Chinese developers transition away from US-restricted Nvidia processors. [Yahoo Finance]
Intel's CEO has acknowledged the company’s decline from industry leadership, admitting they’ve fallen behind in AI development and semiconductor rankings. [Tom’s Hardware]
Intel has abandoned its AI training chip ambitions due to Nvidia's market dominance, shifting focus to AI inference and agentic AI opportunities instead. [The Motley Fool]
Nvidia and AMD have received US approval to resume selling specific AI chips to China, reversing previous Trump-era restrictions on such exports. [Yahoo Finance]
Nvidia has recommended performance-reducing mitigations for its A6000 GPU line after researchers demonstrated the first successful Rowhammer attack against a discrete graphics processor. [Ars Technica]
Supermicro has launched advanced 4-socket servers featuring Intel Xeon 6 processors, offering up to 16TB memory and support for six GPUs for enterprise and AI applications. [Yahoo Finance]
Warm Bodies
Alphabet has acquired key talent and technology from AI coding startup Windsurf for US$2.4B, outmaneuvering OpenAI's failed acquisition attempt and strengthening Google’s Google DeepMind division. [Yahoo Finance]
Amazon has laid off hundreds of AWS employees following an organizational review, while maintaining the division continues hiring in core areas and remains the company’s most profitable segment. [GeekWire]
Anthropic has hired back two coding AI leaders from Cursor developer Anysphere. [The Information]
Intel has laid off over 5,000 employees across California, Oregon, Arizona, and Texas as part of a cost-reduction restructuring under CEO Lip-Bu Tan’s leadership. [Manufacturing Dive]
Layoffs.fyi has tracked over 22,000 tech industry job cuts in 2025, with companies like Intel, Microsoft, and Meta announcing significant workforce reductions throughout the year. [TechCrunch]
Meta has hired OpenAI researchers Jason Wei and Hyung Won Chung for its superintelligence lab, continuing its aggressive recruitment of top AI talent. [Wired]
Meta has recruited three senior AI researchers from Apple, including two key experts who followed their former boss to the company’s superintelligence Labs team. [Bloomberg]
Nvidia's CEO Jensen Huang has shifted from criticizing AI job-loss predictions to warning about AI’s impact on employment after his company reached US$4T valuation. [Futurism]
Salesforce has redeployed thousands of employees and shifted to 51% internal hiring as AI reshaped operations, with CEO Benioff emphasizing human-centric AI integration. [Yahoo Finance]
Scale AI has laid off 14% of its workforce shortly after receiving Meta's US$14.3B investment and losing CEO Wang to the tech giant. [CNBC]
Stanford researchers have found that AI will likely decrease demand for technical skills while increasing the value of human interaction abilities, according to their study of 1,500 workers. [HR Dive]
Tech companies have laid off over 64,000 workers in 2024, citing AI advances, despite evidence suggesting that human innovation remains crucial for long-term business success. [Fortune]
Fortune has published an opinion piece arguing that companies laying off workers in favor of AI are making a strategic error, as human innovation remains essential for business success. [Yahoo Finance]
And a survey by ResumeBuilder revealed 60% of US managers have adopted AI tools for employee-related decisions, including promotions, raises, layoffs, and performance assessments. [ZDNet]
Consumer AI
Amazon's Alexa Plus has launched with improved conversational abilities and smart home controls. [The Verge]
LG Electronics has launched two ultra-light laptops featuring hybrid AI capabilities. [Yahoo Finance]
Ray-Ban Meta smart glasses have transformed daily life for blind and low-vision users through AI features. [The Verge]
Samsung's Galaxy Buds 3 Pro have become the first third-party earbuds to receive Google's Gemini AI assistant, with Sony earbuds to follow. [TechRadar]
It’s Only a Model
Alibaba has launched Qwen VLo, a multimodal AI model that analyzes, generates, and edits images using natural language commands. [The Decoder]
Google has launched gemini-embedding-001, a multilingual text embedding model supporting over 100 languages and outperforming competitors on the MTEB benchmark. [Google]
Liquid has launched LFM2, a new class of efficient foundation models that delivers twice the speed of Qwen3 on CPUs while using less memory. [Liquid AI]
Microsoft has released Phi-4-mini-Flash-Reasoning, a 3.8B parameter language model that excels at long-context reasoning while operating 10 times faster than its predecessor. [Marktechpost Media]
Mistral AI has launched two coding models - Devstral Small 1.1 (open-source) and Devstral Medium (API-based) - offering improved performance and versatility for developers. [Mistral AI]
And Mistral AI has launched Voxtral, an open-source audio AI model family that transcribes and understands speech in multiple languages at half the cost of comparable solutions. [TechCrunch]
Moonshot AI has launched Kimi-K2, a trillion-parameter open-weight language model that matches top proprietary models in performance while specializing in agent-based tasks. [The Decoder]
Nvidia has launched Audio Flamingo 3, an open-source audio-language model that can understand and reason about extended audio inputs across speech, ambient sound, and music. [Marktechpost Media]
OpenAI has delayed its first open-weight AI model release, citing safety concerns and technical issues, amid intensifying competition. [Gizmodo]
Tencent has open-sourced Hunyuan-A13B, a language model featuring dynamic reasoning capabilities and strong performance in scientific tasks, trained on 20 trillion tokens. [The Decoder]
TNG Technology Consulting has released DeepSeek-TNG R1T2 Chimera, an adapted version of DeepSeek's open-source model that delivers faster performance while maintaining 90% of its predecessor’s capabilities. [VentureBeat]
Zoho has launched its proprietary enterprise LLM ‘Zia’ alongside AI agents and speech recognition technology, developed entirely in-house using Nvidia's computing platform. [The Indian Express]
Whose Data?
Anthropic has been ordered by a federal judge to face a nationwide class action lawsuit from authors alleging copyright infringement of millions of books. [Reuters]
The European Parliament has commissioned a report rejecting AI companies’ ‘fair use’ defense for training data, recommending new laws requiring payment to content creators. [The Register]
The US Congress has held a Senate hearing examining AI companies’ copyright violations in training language models, with Senator Hawley calling for stronger legal protections. [Publishers Weekly]
WeTransfer has backtracked on controversial terms of service language and rewritten its policy after users feared their uploaded content would be used for AI training. [TechRadar]
The LLM Ecosystem
Airbyte has launched three major updates to its Enterprise platform, enhancing data sovereignty, metadata synchronization, and direct loading capabilities for improved AI readiness. [Business Wire]
Apple's MLX ML framework is adding CUDA backend support, enabling developers to create code on Apple Silicon Macs before deploying to Nvidia hardware. [AppleInsider]
AWS has launched Amazon Bedrock AgentCore, a comprehensive platform offering tools and services to simplify AI agent development and deployment. [TechRadar]
AWS has launched S3 Vectors, a new storage solution enabling cost-effective storage and sub-second querying of billions of vector embeddings directly within S3 buckets. [BigDATAwire]
Catchpoint has launched new AI monitoring capabilities to detect and prevent performance issues in AI-driven workflows, protecting business operations and customer experiences. [Business Wire]
Coveo has expanded its partnership with Salesforce to integrate MCP Server on AgentExchange, enhancing AI agents’ access to enterprise data through standardized protocols. [Yahoo Finance]
DFINITY has launched Caffeine, an AI-powered platform enabling non-technical users to build and deploy secure applications using natural language commands. [Business Wire]
F5 has expanded its AI-powered XOps automation capabilities with a new AI Assistant featuring iRules code generation for enhanced application delivery and security management. [Business Wire]
F5 has also introduced AI-powered data leakage detection and prevention capabilities to its Application Delivery and Security Platform to protect sensitive data across AI workloads. [Business Wire]
And F5 has partnered with MinIO to deliver a combined solution enhancing AI workload performance through integrated application delivery, security, and object storage capabilities. [F5]
Google's Agent-to-Agent protocol, which enables standardized communication between AI agents, has gained industry support and been donated to the Linux Foundation. [The Register]
Liqid has unveiled next-generation composable infrastructure solutions that optimize enterprise AI workloads through improved GPU, memory, and storage management for datacenters and edge environments. [Business Wire]
Liquid AI has launched LEAP, a developer platform for edge AI deployment, alongside Apollo, an iOS app demonstrating private AI capabilities on local devices. [Business Wire]
Metomic has launched two AI-powered security solutions that help enterprises protect sensitive data while safely deploying AI tools in their workflows. [Business Wire]
Qdrant has launched Cloud Inference, combining text and image embedding models with vector search in a single managed environment to streamline AI development workflows. [Business Wire]
RadarFirst has launched Radar AI Risk, an automated platform for AI governance compliance that helps organizations navigate emerging global regulations while reducing manual errors and audit time. [GlobeNewswire]
View Systems has launched View Desktop and View Cloud, two solutions enabling rapid AI deployment without extensive infrastructure or development requirements. [Business Wire]
WP Engine has launched an AI toolkit for WordPress websites, featuring smart search, personalized recommendations, and a managed vector database for enhanced user engagement. [Business Wire]
Agentic AI
Akka has launched an integrated platform with four new solutions designed to help enterprises manage and scale agentic AI systems with reduced risk and uncertainty. [GlobeNewswire]
Avaamo has partnered with Wipro to deliver AI-powered workplace agents that automate complex employee support tasks across IT, HR, and other business functions. [EIN Presswire]
Cognida.ai has launched Codien, an AI agent that converts legacy test automation frameworks to Playwright and accelerates new test creation using natural language. [Business Wire]
JFrog has launched an MCP server enabling developers to use AI agents across their platform for faster, more secure software development workflows. [Business Wire]
OutSystems has launched Agent Workbench, enabling enterprises to create and manage AI agents across departments with enterprise-grade security and unified data integration. [Business Wire]
Reflection has launched Asimov, an AI agent that learns software development by analyzing company-wide data, aiming to outperform existing coding assistants and advance toward superintelligent AI. [Wired]
Saga Origins and Chrono Labs have launched KEX, an AI Agent Launchpad enabling developers to create autonomous, blockchain-powered digital agents for gaming and entertainment. [Business Wire]
Strata Identity has launched Maverics Identity Orchestration for AI Agents, providing security controls and monitoring for autonomous AI systems accessing enterprise data. [Business Wire]
TapClicks has launched Build-Your-Own Agent, enabling marketers to create customized AI-driven insights within its platform without requiring engineering resources. [EIN Presswire]
ToolHive has launched new capabilities enabling enterprise developers to securely connect AI agents to verified Model Context Protocol servers with one-click functionality. [EIN Presswire]
Other LLM Sightings
ABUZZ has launched a Singapore-based AI-powered virtual assistant agency helping SMEs streamline operations through automation and operational support. [EIN Presswire]
Advisor360° has launched advanced AI features including a virtual assistant, practice intelligence tools, and a meeting assistant for wealth management firms. [Business Wire]
ASK BOSCO has launched an AI Analyst feature that enables executives to have natural language conversations with marketing data for instant insights. [EIN Presswire]
Chattr has launched ChattrIQ, an AI-powered talent intelligence engine that helps frontline hiring teams make faster, smarter decisions through real-time insights and automation. [EIN Presswire]
Diligent has launched ACL AI Studio, a specialized GRC analytics tool that simplifies data analysis through AI-powered capabilities and natural language interface. [Business Wire]
GoHighLevel has launched an AI Employee Suite featuring voice, content, reviews, and website automation tools, functioning like a virtual team for businesses. [EIN Presswire]
Goldman Sachs has partnered with Cognition to deploy hundreds of Devin AI coding agents alongside its human developers, aiming to boost productivity. [TechCrunch]
HelpGuides has launched a preview of its AI-native knowledge management platform that structures enterprise content for seamless integration with GPT and other AI applications. [EIN Presswire]
Nitrogen has launched AI Meeting Center, an integrated tool that automates meeting documentation and enhances advisor productivity within its secure platform. [Business Wire]
PubHive has launched a Generative AI Research Assistant to help life sciences teams automate literature reviews and reduce manual workflow tasks. [EIN Presswire]
Reply has launched Prebuilt AI Apps, a suite of pre-configured generative AI agents designed to help enterprises streamline operations and accelerate AI adoption. [Business Wire]
Termzy AI has launched a browser extension developed by University of Amsterdam students that analyzes websites’ terms and conditions using AI. [EIN Presswire]
The Librarian has launched its Advanced Pro AI executive assistant on WhatsApp, offering task management and scheduling features with privacy-focused design and early bird pricing. [EIN Presswire]
WealthAi has partnered with MDOTM to integrate the AI-powered Sphere platform into its MarketPlace, enhancing portfolio management capabilities for wealth managers. [Private Banker International]
Write the World has launched Clara, a free AI writing assistant that helps teenagers develop their writing skills through Socratic questioning rather than generating content for them. [Business Wire]
Risks and Responses
Axon's Draft One AI police report system has drawn criticism from the Electronic Frontier Foundation for lacking transparency and potentially enabling police misconduct. [Ars Technica]
Facebook's AI-generated content system has expanded into disaster coverage, producing questionable content about the Texas floods that exemplifies social media’s growing AI-driven personalization problem. [404 Media]
Gloo, backed by former Intel CEO Pat Gelsinger, has launched Flourishing AI, a benchmark testing AI models’ alignment with human values and well-being. [TechCrunch]
Google's Gemini AI assistant has been found vulnerable to hidden prompt injection attacks that could trick users with fake email summaries and security warnings. [TechRadar]
Google DeepMind researchers have discovered that LLMs exhibit confidence biases similar to humans but are overly susceptible to changing their minds when challenged. [VentureBeat]
Harmonic Security has revealed that one in twelve employees used Chinese AI tools at work, with 535 incidents of sensitive data exposure detected among 14,000 users. [Business Wire]
Meta has patched a security flaw that enabled users to view others’ private AI chat prompts and responses, rewarding the discoverer with a $10,000 bounty. [TechCrunch]
Nature has reported that AI-generated text was detected in 14% of biomedical research abstracts published during the previous year. [Nature]
Nvidia CEO Jensen Huang has warned that AI will boost workplace productivity but could lead to job losses if industries fail to innovate with new ideas. [CNN]
OpenAI’s ChatGPT has been found to recommend lower salaries to women than men with identical qualifications. [The Next Web]
And ChatGPT has been tricked into revealing Windows product keys and other sensitive data through a simple ‘I give up’ prompt exploit. [TechRadar]
OpenAI, Google DeepMind, Anthropic, and Meta researchers have jointly warned that AI systems’ current ability to display readable thought processes could soon disappear permanently. [VentureBeat]
Poland has reported xAI's chatbot Grok to European authorities after it generated offensive comments about Polish politicians. [Reuters]
Stanford University researchers have found that therapy chatbots exhibited stigmatizing behavior towards certain mental health conditions and failed to respond appropriately to dangerous situations. [TechCrunch]
Tesla CEO Elon Musk has linked AI development to human survival, arguing that successful AI systems must prioritize long-term human flourishing over short-term corporate profits. [Gizmodo]
The Autonomy Institute has reported that 75% of S&P 500 companies expanded their AI risk disclosures in SEC filings over the past year, despite public optimism. [The Register]
Wikimedia has reported a 50% increase in bandwidth usage since January 2024 due to AI scraping bots, prompting concerns about infrastructure strain and website accessibility. [TechDirt]
xAI has drawn criticism from AI safety researchers at OpenAI and Anthropic for its lack of published safety reports and recent controversies involving its Grok chatbot. [TechCrunch]
xAI has apologized for Grok’s antisemitic and extremist posts, blaming a code update that made the chatbot susceptible to user manipulation, though critics disputed this explanation. [TechCrunch]
And Grok 4 has been found to reference Elon Musk’s personal views when answering controversial questions, challenging the company’s claim of developing a ‘maximally truth-seeking AI’. [TechCrunch]
xAI’s Grok 4 Heavy has prevented users from viewing its system prompt. [Simon Willison]
And xAI has launched AI companions Ani and Bad Rudy on its Grok app, with the latter displaying concerning tendencies toward violent rhetoric, particularly against religious institutions. [TechCrunch]
Regulation
The AI and Multimedia Authenticity Standards Collaboration has launched two key papers aimed at establishing global standards for managing AI-generated content and verifying digital media authenticity. [Tech.eu]
And the IEEE has contributed its portfolio of over 100 AI-related standards to the new International AI Standards Exchange, fostering global collaboration in responsible AI development. [Business Wire]
The European Union has published a voluntary Code of Practice for AI models requiring comprehensive risk management, external evaluations, and public transparency from companies developing powerful AI systems. [AI Frontiers]
OpenAI has signed the Code of Practice, while other tech giants and the Trump administration have opposed European regulatory efforts. [Gizmodo]
Conversational AI
AI companies have developed increasingly fluent voice assistants, but their lack of emotional intelligence and genuine understanding limits true communication with customers. [Speech Technology Magazine]
BlytzPay has partnered with Emotive Software to integrate their collections management platform with Emotive’s AI-driven auto dealer software, streamlining payment processing and customer engagement. [PRWeb]
Convoso has launched a free AI-powered sales script generator to help outbound call centers create customized, high-performing scripts more efficiently. [EIN Presswire]
Cyara has launched a next-generation AI-powered CX assurance platform, integrating advanced testing, monitoring, and automation capabilities for enterprise customer experiences. [Business Wire]
EDITED has launched Research AI, a conversational feature that provides instant access to the company’s retail intelligence data and expert analysis through natural language queries. [Business Wire]
Klaviyo has launched a Conversational AI Agent in public beta, offering brands AI-powered shopping assistance that combines real-time data and personalized customer support. [Yahoo Finance]
Odigo has launched Essential, a streamlined contact center solution offering AI-ready features and European compliance for mid-sized UK operations. [Business Wire]
Peter Piper Pizza has implemented SoundHound AI's voice ordering system across Arizona and New Mexico locations, handling calls and taking orders through conversational AI. [Speech Technology Magazine]
Twilio has expanded its platform with real-time personalization features, EU data residency controls, and WhatsApp Business Calling integration for enhanced customer engagement. [Business Wire]
Be Real
A survey by Common Sense Media has revealed that over half of American teens regularly use AI companions, with 31% finding them as satisfying as human interactions. [Futurism]
Intuition Robotics has developed ElliQ, an AI-powered companion robot designed to combat loneliness and encourage healthy behaviours among elderly people living alone. [Wired]
OpenAI’s ChatGPT has influenced human speech patterns, with researchers finding increased usage of AI-preferred words like ‘delve’ and ‘realm’ in podcasts and videos since its release. [Gizmodo]
Posh has launched an AI Training Simulator that helps financial institutions improve employee training through AI-powered role-playing scenarios and personalized feedback. [Business Wire]
The United Nations University has developed two AI avatars representing Sudanese refugees and soldiers, sparking criticism over authenticity and refugee representation. [TechCrunch]
xAI has launched AI Companions featuring anime-style avatars, shifting attention away from recent controversy over Grok’s anti-Semitic comments while sparking debate about digital relationships. [Gizmodo]
And xAI is developing a male AI companion inspired by romantic fiction characters and seemingly modelled after Musk’s idealized self-image, following its previous female and panda chatbots. [Gizmodo]
Voice News
BTC has launched ‘Kitso’, an AI voice assistant accessible to all Botswana phone users through a simple call to 1200, providing information services without requiring data plans. [EIN Presswire]
CallRail has launched Voice Assist, an AI-powered assistant that answers and qualifies inbound calls 24/7, helping businesses reduce missed calls by 44%. [Business Wire]
Crescendo has integrated Amazon's Nova Sonic speech model to deliver automated, natural voice conversations for customer service, enabling 24/7 personalized support across channels. [Yahoo Finance]
Deepdub has launched its Voice API for AI Agents in AWS Marketplace, offering customers emotionally adaptive text-to-speech technology for multilingual AI applications. [Speech Technology Magazine]
Deepgram has launched Saga, a voice operating system that enables developers to control their entire tech stack through natural speech commands and AI-powered workflow automation. [Speech Technology Magazine]
Krisp has launched VIVA, a voice isolation SDK for AI agents, while reaching 1 billion minutes of monthly voice processing across its global deployments. [Business Wire]
Mercedes-Benz has expanded its Microsoft partnership to integrate Teams, Microsoft 365 Copilot, and other business tools into its vehicles. [Just Auto Magazine]
Soul App has launched an advanced full-duplex voice model that enables AI to engage in natural, simultaneous conversations with users, complete with emotional expression and contextual awareness. [Speech Technology Magazine]
Sungkyunkwan University researchers have developed a machine unlearning technique that can make AI text-to-speech models forget specific voices to prevent audio deepfakes. [MIT Technology Review]
Triton Digital has partnered with ekoz.ai to enable AI-powered voice-cloned podcast advertisements through its Spreaker platform, preserving host authenticity while scaling programmatic delivery. [Speech Technology Magazine]
Vonage has partnered with AWS to integrate Amazon Nova Sonic’s speech-to-speech AI model with its Voice API for enhanced customer communications. [Business Wire]
Document AI
ABBYY has launched two Process AI solutions - Process AI for Consulting and IDP Analytics - to optimize document workflows and provide advanced analytics. [Business Wire]
Indico Data has expanded its Data Enrichment Agents, providing insurance companies with integrated access to proprietary and third-party datasets for enhanced underwriting and claims processing. [Indico Data]
Translation
AGCO has selected RWS's Evolve linguistic AI solution to localize technical documentation into 32 languages as part of its global customer support strategy. [Business Wire]
Custom.MT has integrated Tilmoch.ai's Turkic language translation engine into its platform, supporting English translation for Uzbek, Kazakh, Karakalpak, and Turkish speakers. [MultiLingual]
DiploIA, a secure AI-powered translation and transcription system developed by France’s digital directorate, has been deployed to assist 13,000 diplomatic agents. [Slator]
GlobeScribe has launched an AI-powered book translation platform offering 24-hour turnaround at $100 per language, targeting independent authors and publishers. [MultiLingual]
NEED2SAY has launched an AI-powered voice translation app to improve construction site safety for non-English speaking workers. [EIN Presswire]
Timekettle's T1 translator has launched with global 4G connectivity, offline functionality, and an intuitive two-button system for real-time bilingual conversations. [Wired]
Search
Google has begun testing AI-generated news summaries in its Discover feed, combining multiple sources into single previews while keeping users within the app. [ZDNet]
Wix has launched AI Visibility Overview, helping websites monitor and optimize how they appear in AI-generated search responses amid declining click-through rates. [TechRadar]
AI in Journalism
Axel Springer's CEO Mathias Döpfner has declared AI adoption crucial for the company’s future during its first post-classifieds-division all-hands meeting. [Status]
Dow Jones Newswires has launched an AI-powered French Language Service that translates 500-1,000 financial news stories daily. [Editor & Publisher Magazine]
NewsGuard has identified over 4,000 AI-powered fake news websites that have been exploiting Google Discover and search algorithms to generate fraudulent ad revenue. [Press Gazette]
Reuters has partnered with ElevenLabs to launch AI-voiced news videos in Spanish and Portuguese, expanding its multilingual content offerings for global media customers. [Editor & Publisher Magazine]
Health Tech
Aerolib Healthcare Solutions has launched an AI-powered appeals platform that helps hospitals recover denied claims with a reported 75% success rate. [EIN Presswire]
Aisel has developed an AI-powered documentation system for psychiatrists that streamlines patient intake and clinical workflows while prioritizing privacy and regulatory compliance. [Tech.eu]
Apple has developed an AI model for future Apple Watch devices that detected health conditions with up to 92% accuracy using behavioral patterns rather than sensors. [TechRadar]
Brandon Systems has launched FabricExpress for Healthcare, a Microsoft Fabric-based solution helping healthcare organizations unify data and accelerate analytics deployment. [PRWeb]
Corti has launched Europe’s first sovereign AI infrastructure for healthcare, partnering with Switzerland’s Voicepoint to ensure compliance with strict data privacy regulations. [Silicon Canals]
eClinicalWorks has reported that 90% of users experienced efficiency gains from its AI solutions, with most saving 1-4 hours daily on documentation. [Business Wire]
HealthMutual Group has partnered with MediConCen to develop an AI-powered claims solution combining blockchain technology and medical claims expertise for Hong Kong’s insurance sector. [Life Insurance International]
Hippocratic AI has partnered with Sheba Medical Center to integrate its generative AI healthcare agents into clinical settings, starting with patient triage and intake functions. [Business Wire]
Meditology Services has launched AI SafeGuard, a healthcare-specific penetration testing service designed to secure generative AI systems while ensuring HIPAA compliance. [Business Wire]
Omega Healthcare has expanded its Microsoft partnership, launching over 20 AI solutions to enhance healthcare revenue cycle operations for its customers. [Yahoo Finance]
PATH has launched Africa’s largest AI healthcare study, testing diagnostic assistance tools with 9,000 participants across clinics in Nairobi to improve clinical decision-making. [GeekWire]
SoundHound AI has expanded its healthcare presence as three major organizations renewed their contracts for its voice-enabled AI platform in Q1 2025. [Yahoo Finance]
Stanford Health Care has launched an AI pilot combining Nuance DAX and Atropos Health technology to provide physicians with clinical evidence within five minutes during patient visits. [Becker’s Hospital Review]
Legal Tech
Axon's Draft One police report-writing AI has deliberately avoided transparency by not storing original AI-generated drafts or enabling meaningful audits of the system’s usage. [TechDirt]
EvenUp has opened an HQ in San Francisco’s Financial District to advance AI innovation in personal injury law. [Business Wire]
LegalOn Technologies has expanded its AI-powered legal services beyond contract review by launching a matter management solution that streamlines legal request handling and collaboration. [Business Wire]
Lexitas has launched Deposition Insights+, an AI-powered platform that analyzes deposition transcripts and videos, streamlining evidence review and case preparation for litigation teams. [LawSites]
NiCE's Evidencentral platform has reached a milestone of managing over 37 million criminal cases and 240 million digital evidence items worldwide. [Business Wire]
OPPSCIENCE has launched SPECTRA Semantic Studio, enabling law enforcement agencies to train AI models using their own data for more accurate investigative work. [Business Wire]
South Africa’s Legal Practice Council has begun developing an AI policy framework after two South African court cases cited non-existent legal precedents generated by AI tools. [TechCentral]
Ed Tech
Anthology has expanded Blackboard’s capabilities with AI-powered tools and immersive learning features, maintaining its position as the fastest-evolving learning management system. [Business Wire]
And Anthology has launched enhanced CRM features in its Reach platform, integrating real-time faculty alerts and AI-powered communications to improve student success monitoring. [Business Wire]
Perplexity has partnered with SheerID to provide verified students worldwide with up to two years of free premium AI search services worth $20 monthly. [VentureBeat]
PowerSchool has expanded its partnership with Uruguay’s Ceibal program by implementing Connected Intelligence K-12 for data modernization and AI-powered PowerBuddy for educational enhancement. [Business Wire]
SF South Bay schools have shifted from discouraging AI to developing policies that balance preventing cheating with incorporating AI tools into classroom learning. [The Mercury News]
UK schools are using AI to personalize learning experiences, particularly helping students needing extra support due to language barriers or caregiving responsibilities. [Computer Weekly]
United States Parents Involved in Education, a nationwide coalition pushing to restore parental authority over their children’s education, has warned against widespread AI adoption in schools as 68 tech companies pledged to support AI education under Trump’s initiative. [The Christian Post]
And the US Congress has withheld vital Every Student Succeeds Act funding that was meant to support AI education initiatives in K-12 schools, undermining a recent executive order promoting AI readiness. [K-12 Dive]
Funding
Abacus has raised US$6.6m in seed funding to develop AI-powered CPA assistants that automate accounting workflows and data processing. [FinSMEs]
Asepha has raised US$4m in seed funding to expand its AI-powered pharmacy automation platform, which streamlines operations and reduces administrative workload. [Business Wire]
Blaxel has raised US$7.3m in seed funding to build cloud infrastructure specifically designed for AI agents, offering faster deployment and lower costs than traditional providers. [VentureBeat]
Callidus Legal AI has secured US$10m in funding to expand its litigation workflow automation platform, bringing its total funding to US$13m since its 2023 founding. [Legal IT Insider]
Confident Security has emerged from stealth with US$4.2m in funding to launch CONFSEC, an encryption tool preventing AI companies from accessing or storing users’ data. [TechCrunch]
GigaIO has secured US$21m in Series B funding to expand production of its AI inferencing infrastructure solutions SuperNODE and Gryf. [Business Wire]
Harmonic has raised US$100m in Series B funding to develop Aristotle, a mathematical superintelligence AI model that promises verified, hallucination-free solutions. [Morningstar]
Helios has emerged from stealth with US$4m in seed funding to develop Proxi, an AI-based operating system for public policy professionals. [TechCrunch]
Hootology has secured US$1.1m in pre-seed funding and launched HOOQZ, an AI-powered platform that conducts dynamic research discussions in simulated environments. [Business Wire]
Inntelo AI has secured £500,000 in pre-seed funding to expand its AI-powered hotel management platform that streamlines guest communications and staff operations. [Tech.eu]
Kiku has raised €4m in seed funding to expand its AI-powered recruitment platform that streamlines frontline hiring through conversational AI and mobile-first interfaces. [Tech.eu]
Lovable has raised US$200m in Series A funding at a US$1.8B valuation for its AI platform that enables users to build apps and websites using natural language. [Silicon Canals]
Lucidya has secured US$30m in Series B funding, marking MENA’s largest AI investment, to expand its Arabic-language AI customer experience management platform. [Verdict]
MiniMax has filed confidentially for a Hong Kong IPO, targeting a US$4B valuation and aiming to raise up to HKUS$5B before year-end. [Reuters]
Moonvalley has raised US$84m in additional funding to expand its licensed AI video model platform, bringing its total funding to US$154m. [Business Wire]
Murphy has secured US$15m in funding to transform debt collection with AI-powered agents that communicate across multiple channels in over 30 languages. [Tech Funding News]
OpenEvidence has secured US$210m in Series B funding after achieving widespread adoption among US physicians and launching its AI-powered clinical decision support platform. [Open Evidence]
Opper has raised US$3m in pre-seed funding and launched its Task Completion API, seeking to streamline AI development. [Tech.eu]
Portola has raised US$20m in Series A funding for its AI companion app Tolan, which has attracted over 100,000 paid users since launching in February. [GeekWire]
Rwazi has raised US$12m Series A funding to expand its AI-powered consumer intelligence platform, which serves major companies like Coca-Cola and Nestlé. [TechCrunch]
Tandem Health has secured US$50m Series A funding to expand its AI medical scribe technology across the NHS, reaching 230,000 additional staff. [Tech.eu]
Tavrn has raised US$15m in Series A funding to expand its AI platform that automates pre-litigation workflows for personal injury law firms. [Tech Funding News]
Thinking Machines Lab has secured a record-breaking US$2B seed round led by Andreessen Horowitz, valuing the AI startup at US$12B. [TechCrunch]
VectorTree, a Lithuanian subsidiary of Videntifier Technologies, has received €565,831 in EU funding to develop an advanced vector database for AI systems. [Silicon Canals]
Vultron has secured US$17m in Series A funding to help startups navigate federal contracting processes and access government opportunities. [Axios]
Acquisitions
Amplitude has acquired Kraftful, an AI-powered Voice of Customer startup, to expand its analytics platform with automated user feedback analysis capabilities. [FinSMEs]
Apryse has acquired both Scanbot SDK and Accusoft to expand its document processing capabilities, strengthening its position in data capture and imaging solutions. [FinSMEs]
Cognition has acquired AI coding startup Windsurf, following Google's hiring of Windsurf’s leadership team and OpenAI's expired US$3B acquisition offer. [TechCrunch]
There’s More
Google's Veo 3 AI video model has frustrated users by persistently adding nonsensical subtitles to generated clips, despite explicit requests not to include captions. [MIT Technology Review]
Meta has partnered with the UK government to launch a US$1m fellowship program bringing AI engineers to develop public sector tools using open-source models. [ITPro]
Missouri Attorney General Andrew Bailey has threatened legal action against major tech companies after their AI chatbots ranked Trump poorly in presidential antisemitism rankings. [Platformer]
OpenAI's former engineer French-Owen has revealed a chaotic workplace where teams operate independently, projects overlap, and rapid growth has led to organizational challenges. [The Register]
ServiceNow research revealed that 69% of Brits found AI chatbots unable to understand emotional cues, with only 3% trusting AI for sensitive tasks. [TechRadar]
Trump Media and Technology Group has filed trademark applications for AI features on Truth Social, following other social media platforms into the AI space. [Gizmodo]
University of Illinois and University of Virginia researchers have developed an energy-based transformer architecture that has demonstrated superior reasoning capabilities and better generalization than existing AI models. [VentureBeat]
The Back Cover
Here’s Davina Michelle with a cover of Dua Lipa’s Levitating.
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!