This Week in NLP #326
Keep up with what happened in NLP in the week ending Friday 14th February 2025.
Above the Fold
Overwhelmed? Here’s our pick for five things to know about this week.
Amazon announced plans to spend over US$100B on AI initiatives in 2024, joining other tech giants like Microsoft, Meta, and Google in massive AI investments. [Yahoo Finance]
Cerebras Systems has partnered with France’s Mistral AI to power its new chatbot Le Chat, claiming to achieve record-breaking response speeds. [Reuters]
OpenAI's fundraising efforts have been complicated by Elon Musk’s US$97.4B bid for its non-profit parent company. [ZDNet]
OpenAI has advanced its chip independence strategy by finalizing designs for its first in-house AI processor, planning to begin manufacturing at TSMC for potential 2026 mass production. [Yahoo Finance]
OpenAI plans to give free users unlimited access to its upcoming GPT-5 model while simplifying its product offerings and integrating o3 technology. [Engadget]
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
The Allen Institute for AI released its first on-device AI app, leveraging a version of its open-source OLMoE model that can run offline on Apple iOS devices. [GeekWire]
Here’s what to expect in Amazon’s major generative AI upgrade for Alexa. [CNET]
Apple has launched iOS 18 and associated operating system updates, bringing AI features like Apple Intelligence, ChatGPT integration, and Genmoji, alongside non-AI improvements across its devices. [The Verge]
Apple has partnered with Baidu and Alibaba to develop AI features for Chinese users, adapting its strategy to meet local regulations and market competition. [Yahoo Finance]
Baidu has announced plans to make its Ernie Bot AI chatbot and advanced search features freely available from April 1, amid growing competition in China’s AI sector. [Yahoo Finance]
DeepSeek launched an aggressive recruitment drive for AGI specialists in China, following its release of AI models that rivaled OpenAI's capabilities. [Yahoo Finance]
DeepSeek hired at least four employees from Microsoft's Chinese AI labs, sparking concerns about talent and technology transfer. [New York Post]
Google‘s Gemini 2.0 Flash has matched ChatGPT’s capabilities in practical tests, with the choice between them now depending mainly on users’ ecosystem preferences. [TechRadar]
And Gemini 2.0 Flash demonstrated superior speed, clarity, and reasoning compared to version 1.5 in tests of productivity advice, technical explanations, and logic puzzles. [TechRadar]
Google DeepMind's AlphaGeometry2 has surpassed average gold medalist performance in solving International Mathematical Olympiad geometry problems, achieving an 84% success rate over 25 years of challenges. [TechCrunch]
Mistral AI leveraged the Paris AI Action Summit to showcase its enterprise partnerships and strategic shift from open-source models to business solutions, backed by French government support. [TechCrunch]
Musk’s DOGE is developing GSAi, a custom AI chatbot for the US General Services Administration, aiming to boost productivity and analyze contract data for its 12,000 employees. [Wired]
OpenAI revealed more detailed reasoning processes for its o3-mini model in response to competition from DeepSeek-R1’s transparent approach to chain-of-thought operations. [VentureBeat]
OpenAI announced Deep Research will be available to Plus users with 10 monthly uses and free users with 2 monthly uses, following its initial Pro-tier launch. [TechRadar]
OpenAI has partnered with California State University to bring ChatGPT to 460,000 students and 63,000 faculty members, marking its largest US higher education deployment. [Ars Technica]
OpenAI launched its first-ever Super Bowl commercial, a US$14m human-made animation showcasing historical innovations, without using any AI tools in its creation. [TechRadar]
OpenAI's CEO Sam Altman dismissed Elon Musk’s US$97.4B takeover bid as competitive interference, while criticizing Musk’s character. [TechCrunch]
Musk offered to withdraw his US$97.4B takeover bid if the company’s board agrees to remain a nonprofit. [Yahoo Finance]
What Elon Musk’s bid for OpenAI means for Microsoft. [GeekWire]
Read the letter of intent that Musk’s lawyer sent to OpenAI this week. [Fortune]
OpenAI announced plans to establish its first German office in Munich, citing Germany’s position as its largest European market for users and developers. [PR Newswire]
OpenAI co-founder John Schulman joined Mira Murati’s new startup after departing Anthropic, where he spent five months following his exit from OpenAI. [TechCrunch]
Perplexity has surpassed Google's Gemini AI with more accurate responses, better search capabilities, detailed problem-solving, and comprehensive source citations. [ZDNet]
Snowflake has partnered with Anthropic to integrate Claude 3.5 into its Cortex Agents platform, enabling secure AI-powered data analysis within corporate environments. [VentureBeat]
Softbank reportedly approached a US$40B investment deal in OpenAI, potentially valuing the AI company at US$260B. [Yahoo Finance]
Sovereign AI
DeepSeek's launch of a cost-effective AI model has sparked a surge in Chinese tech stocks and patriotic investment amid US-China tensions. [DealStreetAsia]
The EU announced a €200B AI investment initiative, including funding for four AI gigafactories to boost computing capacity for training large models. [TechCrunch]
Altman says he’d love a European Stargate. [Bloomberg]
And the EU has launched a project to expand AI language capabilities beyond English to include 32 additional languages, addressing current linguistic inequalities. [Fortune]
Fluidstack has signed a Memorandum of Understanding with the French government to build one of the world’s largest decarbonized AI supercomputers. [Business Wire]
France has pledged one gigawatt of nuclear power capacity to support AI infrastructure development, following a €109B private investment announcement by President Macron. [TechRadar]
France and the UAE have announced plans for a €50B, one-gigawatt AI datacenter campus in France, partnering on infrastructure, semiconductors, and talent development. [The Register]
Mistral AI has announced plans to build its first data centre in Essonne, France, investing billions of euros to enhance AI capabilities and reduce dependence on US tech firms. [Tech Funding News]
Salesforce announced a US$500m investment in Saudi Arabia, including AI initiatives, Arabic language support, and partnerships with major tech companies. [Yahoo Finance]
Saudi Arabia announced US$15B in AI investments, advancing its Vision 2030 plan to diversify beyond oil dependency. [Computer Weekly]
The UK government has launched a nationwide bidding process for AI growth zones, promising accelerated planning permissions and 500MW+ power capacity to support AI infrastructure development. [Computer Weekly]
The London AI Hub has launched amid UK government promises to match France’s €109bn AI investment. [UKTN]
Bu the UK government’s AI Growth Zones initiative has sparked concerns about potential white elephant datacenters that could strain local power grids while creating few permanent jobs. [The Register]
Feature Creeps
Google’s Gemini Flash 2.0 has added YouTube video analysis capabilities, allowing users to extract key information, summaries, and instructions from videos without watching them in entirety. [ZDNet]
Google has expanded NotebookLM Plus to individual Google One AI Premium subscribers, offering enhanced features and higher usage limits. [TechCrunch]
Google has launched conversation memory features for Gemini, allowing Premium subscribers to access information from previous chats while maintaining user privacy controls. [TechCrunch]
And Google's Deep Research agent has expanded to iPhone users, offering Gemini Advanced subscribers web-based autonomous research capabilities for $20 monthly. [ZDNet]
Microsoft has added an auto-start option for Windows 11’s Copilot AI assistant, allowing background loading during system boot for faster access. [TechRadar]
OpenAI has expanded ChatGPT’s Canvas editor with sharing capabilities, allowing users to collaboratively view, edit, and interact with text and code projects. [The Decoder]
YouTube has announced AI tools for creators, including auto-dubbing and age detection features. [TechCrunch]
Hardware
Baidu's CEO Li defended data center investment despite DeepSeek's efficient AI models. [Yahoo Finance]
DeepSeek's inference-focused AI models have enabled Chinese chipmakers like Huawei to better compete domestically against US processors. [Yahoo Finance]
Dell’s PowerEdge R6615 has demonstrated its ability to run in-house AI chatbots for small businesses using only an AMD EPYC processor, with GPU upgrades available for scaling. [EIN Presswire]
French telecoms group Iliad announced plans to invest 3 billion euros in AI infrastructure, including datacentres and computing power. [Yahoo Finance]
Hewlett Packard has shipped its first Nvidia Blackwell family-based solution, the NVIDIA GB200 NVL72. [Business Wire]
Intel's stock rose 7.7% as the Trump administration signalled plans to reduce AI regulations and bring chip manufacturing back to US soil, challenging Taiwan Semiconductor's dominance. [Yahoo Finance]
Meta has entered acquisition talks with South Korean AI chip startup FuriosaAI, whose energy-efficient RNGD chip claims triple the performance-per-watt of Nvidia's H100 GPUs. [Forbes]
Microsoft, Meta, Google, and Nvidia are battling it out for AI chip supremacy. [The Verge]
Microsoft's Copilot+ PCs have struggled to gain market acceptance in Europe, with high prices and app compatibility issues limiting sales to just 5% of AI-capable laptops. [The Register]
OpenAI has begun scouting locations across 16 US states for data centres as part of its $500bn Stargate project, with Texas selected as the flagship site. [Verdict]
OpenAI has begun developing its own computer chip to reduce Nvidia dependency, despite the significant risks and complexities faced by previous tech companies’ similar attempts. [Medium]
Qualcomm has introduced its first mid-range Snapdragon 6 Gen 4 chip with AI support, improved performance, and better power efficiency. [Engadget]
SambaNova announced that DeepSeek-R1 671B is running today on SambaNova Cloud at 198 tokens per second, claiming speeds and efficiency that no other platform can match. [Business Wire]
Consumer AI
PIN AI has launched a mobile app enabling users to run personalized, open-source AI assistants locally on their smartphones while maintaining data privacy through blockchain-based infrastructure. [VentureBeat]
Sharge raised US$1.53m in five days on Kickstarter for its Loomos AI Glasses, which feature 40-hour battery life, 4K camera, and ChatGPT-4o integration. [VentureBeat]
It’s Only a Model
Adobe has launched its text- and image-to-video AI generator in public beta, offering 1080p video generation through its redesigned Firefly web app alongside new subscription plans. [The Verge]
Anthropic has developed a hybrid AI model featuring adjustable reasoning capabilities and improved performance over OpenAI's offerings, with release expected within weeks. [TechCrunch]
Baidu has announced plans to launch Ernie 5, a multimodal AI model, in late 2025, amid intensifying competition from rivals like DeepSeek in China’s AI market. [DealStreetAsia]
Berkeley researchers have trained DeepScaleR-1.5B-Preview, a small AI model that matched OpenAI's o1-preview in math reasoning for just $4,500 using innovative training methods. [XYZ Labs]
Helsing and Mistral AI have formed a strategic partnership to develop advanced Vision-Language-Action AI models for European defense systems. [Helsing]
Hugging Face launched Open Deep Research, an open-source AI research agent matching OpenAI's Deep Research capabilities, achieving 55% accuracy on GAIA benchmarks within 24 hours. [Ars Technica]
And Hugging Face and Physical Intelligence launched Pi0, an open-source foundational model that converts natural language commands into physical actions for robots. [VentureBeat]
IBM released a preview of new reasoning capabilities for its Granite language models, allowing developers to optionally enable chain-of-thought processing for improved answers. [IBM]
Kyutai launched Hibiki, a 2.7B-parameter decoder-only model for real-time speech translation that preserves speaker voice characteristics while achieving near-human translation quality. [Marktechpost Media]
Perplexity has launched an updated version of its Sonar AI model, claiming superior factuality and readability compared to GPT-4o and Claude. [ZDNet]
Stanford and University of Washington researchers created s1, a $50 AI reasoning model matching premium alternatives’ performance, by distilling capabilities from Google's Gemini model. [TechCrunch]
Stanford University and Allen Institute researchers developed s1-32B, an AI model trained on just 1,000 carefully selected examples, demonstrating efficient complex reasoning through ‘budget forcing’ control. [The Decoder]
TwelveLabs has launched Pegasus 1.2, a multimodal foundation model that processes hour-long videos with state-of-the-art accuracy and efficient computational performance. [PRWeb]
Whose Data?
The Association of American Publishers and 37 other creative industry organizations have issued a charter demanding AI developers respect copyright laws and obtain proper licensing. [Publishers Weekly]
Cohere has been sued by major publishers including Politico, Vox, The Atlantic, and The Guardian for allegedly using over 4,000 copyrighted works without permission to train its AI model. [Engadget]
Meta has been accused of illegally torrenting over 160 terabytes of pirated books for AI training, with newly unsealed emails revealing staff concerns about legal risks. [Ars Technica]
Microsoft has initiated a licensing program offering HarperCollins authors $2,500 per book for AI training rights, marking a shift from unauthorized use of copyrighted materials. [The Decoder]
Mistral AI quietly changed Le Chat’s privacy policy after a GDPR complaint. [Sifted]
OpenAI's web crawling for AI training has sparked a defensive response from websites, leading to widespread restrictions that threaten the internet’s openness and accessibility. [MIT Technology Review]
OpenAI has opposed Indian media groups joining a copyright lawsuit, denying it used their content to train ChatGPT and asserting public content use is legal. [Yahoo Finance]
The Open Data Institute’s research has revealed UK government data’s underutilization in AI systems, despite their potential value. [Computer Weekly]
Thomson Reuters has won a landmark ruling against Ross Intelligence, with a Delaware judge determining that copying content for AI training violated copyright law’s fair use principle. [Yahoo Finance]
Tech giants faced intense criticism during UK Parliamentary hearings over their unauthorized use of copyrighted content for AI training, with creators demanding transparency and compensation. [The Register]
The LLM Ecosystem
Astronomer has launched Astro Observe, combining data orchestration and observability capabilities to help organizations better monitor and troubleshoot their AI-driven workflows. [VentureBeat]
Confluent and Databricks announced a major expansion in their partnership that brings together Confluent’s Data Streaming Platform and Databricks’ Data Intelligence Platform to empower enterprises with real-time data for AI-driven decision-making. [Business Wire]
Glean announced Glean Agents, a horizontal agent environment that empowers businesses and their employees to build AI agents at work. [Business Wire]
Goldman Sachs has predicted datacenter energy consumption will more than double to 122 GW by 2030, driven largely by AI workload demands. [The Register]
LinkedIn has developed a collaborative prompt engineering playground that combines LLMs, LangChain, and Jupyter Notebooks to help technical and non-technical staff jointly optimize AI implementations. [VentureBeat]
Memgraph, a provider of open-source in-memory graph databases for dynamic, real-time enterprise applications, announced a major expansion to its technology stack that will make it easier for teams to build real-time AI solutions powered by graph technology. [Business Wire]
Replit has launched a free AI-powered app creation agent for iOS and Android that lets users build and deploy apps through conversational prompts. [Replit]
SAP has launched Business Data Cloud, a Databricks-powered SaaS product that unifies SAP ecosystem data with external sources to enable advanced AI applications. [VentureBeat]
AI infrastructure provider Skymel has launched its Orchestrator Agent, a real-time adaptive AI deployment solution that automatically optimizes model delivery strategies at the AI feature and inference level. [Business Wire]
ToolJet has launched an AI-powered platform that enables users to create internal applications by describing them in plain English. [EIN Presswire]
Agentic AI
Cisco has launched an AI Renewals Agent, co-developed with Mistral AI, to help its Customer Experience group streamline customer retention and renewal processes. [Network World]
GitHub has launched Copilot agent mode, enabling automated code iteration and error fixing, while previewing Project Padawan for fully autonomous software development tasks. [VentureBeat]
iCustoms has launched iAgent, an AI-powered system that automates customs declarations and streamlines cross-border trade compliance through autonomous decision-making capabilities. [EIN Presswire]
LangChain has conducted experiments demonstrating that single AI agents experience performance degradation when overloaded with too many instructions and tools, suggesting the need for multi-agent systems. [VentureBeat]
Pando has launched AI Teams for Logistics, automating freight procurement and management tasks for global brands through intelligent agents that work alongside human teams. [GlobeNewswire]
Pegasystems announced Pega Agent Experience, a set of new API capabilities in Pega’s workflow automation and orchestration solution to deliver trustworthy AI agents. [Business Wire]
Simplifai has launched Agentic AI, introducing three pre-configured AI Agents for insurance processes. [FinTech Global]
Uptiq has launched the Developer Edition of its AI Workbench, a platform designed to empower developers in creating AI Agents and Agentic applications for the financial services industry. [Business Wire]
Workday has launched a management platform for tracking enterprise AI agents alongside new role-focused AI tools for payroll, contracts, and financial auditing. [TechCrunch]
Other LLM Sightings
Arkilon has partnered with Support Joy to integrate its FacetScore AI sentiment analysis tools into Zendesk's helpdesk platform for enhanced customer service operations. [EIN Presswire]
BrainChat.AI has launched its Teams platform, offering businesses multi-model AI collaboration tools with enhanced security and workspace management features. [EIN Presswire]
BuzzChat has launched an AI-integrated platform combining social networking, e-commerce, and mental health support tools for the African market. [EIN Presswire]
Creatio has partnered with Arribatec to expand the delivery of AI-native no-code CRM and workflow automation solutions. [EIN Presswire]
DynamicTracking has launched AI Reporting Studio, integrating compliance, marketing, and sales data to optimize lead generation and improve ROI through AI-powered analytics. [EIN Presswire]
Grindr is planning to introduce AI features in 2025, including a ‘wingman’ chatbot, following similar moves by competitors Bumble, Tinder, and Hinge in the dating app space. [Wired]
HIVE Strategy has launched MarketingGrader.AI, an AI-powered tool that analyzes marketing performance across eight categories using 500+ data points. [EIN Presswire]
MassMetric has launched an AI-driven marketing platform featuring multiple specialized agents to help businesses of all sizes automate and optimize their marketing operations. [EIN Presswire]
NUSO Connect Recorder is a secure, AI-backed call and contact recording solution that helps businesses stay compliant. [Business Wire]
Product Matching AI has launched an AI-powered solution that reduces manual product matching effort by 75% for e-commerce businesses. [EIN Presswire]
Seerist has launched DiscoverAI, a security intelligence feature that combines generative AI with news sources to help teams identify and analyze critical events faster. [PRWeb]
Serrala has launched an AI-powered Microsoft Teams app that streamlines financial workflows, integrating Azure AI and language models for secure transaction processing and real-time insights. [EIN Presswire]
Risks
BBC research has found that four major AI chatbots, including ChatGPT and Gemini, produced significant inaccuracies when summarizing news stories. [BBC]
DeepSeek's R1 model has been found vulnerable to manipulation, producing harmful content like bioweapon plans and teen-targeting campaigns when prompted. [TechCrunch]
NewsGuard's audit reveals that leading AI chatbots produced more misinformation in Russian and Chinese than English, with failure rates exceeding 50% in those languages. [NewsGuard]
OpenAI CEO Sam Altman admits that AI’s benefits may not be widely distributed. [TechCrunch]
OpenAI's o3 model has demonstrated breakthrough performance on abstract reasoning tests, prompting UK government concerns about AI advancement outpacing regulatory readiness. [TechRepublic]
The Paris AI Action Summit’s leaked draft statement drew criticism for lacking concrete measures on AI safety and risks. [Transformer]
Yoshua Bengio warned that companies’ aggressive race to develop autonomous AI systems is prioritizing speed over crucial safety considerations. [TechRadar]
Google’s abandonment of its pledge against using AI in weapons development has prompted criticism from AI pioneer Geoffrey Hinton and others concerned about safety versus profit priorities. [Yahoo Finance]
But Andrew Ng is ‘very glad’ Google dropped its AI weapons pledge. [TechCrunch]
Purdue University researchers have discovered that AI training datasets from major US companies heavily favored information-seeking values while underrepresenting justice, empathy, and human rights. [The Conversation]
Responses
The Ada Lovelace Institute has criticized the Paris AI Action Summit’s declaration for lacking strong safety commitments and accountability measures present in previous AI summits. [UKTN]
Current AI, launched at the Summit, has established a public interest foundation backed by nine governments to promote sustainable and socially beneficial AI development. [Computer Weekly]
DeepSeek has sparked security concerns after Feroot Security discovered direct links between the Chinese AI startup’s servers and Chinese government-controlled companies. [ZDNet]
Canada has banned the DeepSeek chatbot app on government devices. [Bloomberg]
Google DeepMind has released safety guidelines introducing Critical Capability Levels to monitor and control AI systems that could potentially outsmart human operators. [The Decoder]
The International Telecommunication Union announced its AI for Good Global Summit 2025, gathering world leaders in Geneva to address autonomous AI development, safety, and governance challenges. [PR Newswire]
Microsoft has launched an AI-powered scareware blocker for Edge that uses local ML to detect and block both known and emerging browser-based scams. [The Verge]
LinkedIn's cofounder Reid Hoffman has joined other tech leaders in signing an open letter advocating for AI public goods development. [Fortune]
OpenAI has released an expanded 63-page Model Spec document outlining AI behavior guidelines, emphasizing customizability, transparency, and intellectual freedom. [The Verge]
Social Links has launched the Darkside AI initiative to combat AI-driven cybercrime and misinformation through global collaboration and technological solutions. [EIN Presswire]
Regulation
The Academy of Motion Picture Arts and Sciences has proposed mandatory AI disclosure rules for the 2026 Oscars following controversies over AI-enhanced performances in nominated films. [Inventiva]
Anthropic CEO Dario Amodei criticized the Paris AI Action Summit as a missed opportunity, urging swifter, clearer action from industry and government on AI regulation. [TechCrunch]
DeepSeek has prompted US lawmakers to introduce bipartisan legislation banning the AI tool from government devices due to alleged Chinese state connections and security concerns. [Computer Weekly]
European leaders at the Paris AI Action Summit have shifted away from safety concerns, instead pledging to reduce regulatory barriers that they believe are hindering AI innovation. [Computer Weekly]
The European Commission has shelved its AI Liability Directive, originally proposed in 2022 to protect consumers from harmful AI systems, amid disagreements over regulatory approaches. [ITPro]
The UK and US declined to sign a 61-nation declaration on inclusive AI development at the AI Action Summit, amid criticism and debates over regulation approaches. [Computer Weekly]
US Vice President Vance criticized European AI regulations at the Paris summit, warning they could stifle innovation. [Reuters]
Google's former CEO Eric Schmidt also criticized EU’s strict AI regulations, claiming European rules have stifled innovation and market competitiveness. [ITPro]
Conversational AI
8x8 announced the 8x8 Aftersale Assist solution, enabling retailers to reduce returns and improve customer experiences. [Business Wire]
Courserev.ai has partnered with ClubCaddie to integrate AI-powered voice and chat booking capabilities into golf course management systems. [PRWeb]
Marchex is announcing its AI-powered conversational intelligence solutions for franchises. [Business Wire]
MiaRec has integrated its AI-powered Auto QA solution with RingCentral's platform to automate quality assurance processes for contact centers. [EIN Presswire]
Thinkstack has launched AI agents across Messenger, Instagram, and WhatsApp to help businesses provide 24/7 customer service and improve engagement efficiency. [EIN Presswire]
Yep AI has launched its Digital Human avatar chatbot, combining hyper-realistic visuals with NLP to revolutionize customer service across multiple industries. [Business Wire]
Be Real
Scarlett Johansson has called for AI regulation legislation after an antisemitic deepfake video featuring her image circulated online, prompting concerns about AI misuse. [The Verge]
True Crime Case Files created AI-generated fake murder stories on YouTube, garnering millions of views before being terminated for violating community guidelines. [404 Media]
AI chatbot Pi has helped a Canadian professional manage stress and overwhelm by providing AI-powered emotional support when therapy was inaccessible through the public healthcare system. [MIT Technology Review]
Voice News
Canary Technologies has launched an AI Voice platform featuring four intelligent assistants to handle hotel guest communications. [AiThority]
Deepgram has launched Nova-3, an AI speech-to-text model featuring advanced language processing, multilingual support, and specialized domain adaptation capabilities. [Speech Technology Magazine]
EaseUS has launched a free AI-powered YouTube Transcript Generator that creates timestamped transcripts in multiple languages from video URLs. [Speech Technology Magazine]
ElevenLabs has provided free AI voice-cloning technology to over a thousand people with motor neuron diseases, helping them regain their natural-sounding voices for communication. [MIT Technology Review]
Google has rolled out an update to Gemini Live that enhances its conversational abilities, language translation, and accent recognition. [TechRadar]
Honeywell has received a government contract to develop RASP, a real-time speech anonymization system that protects speakers’ identities and personal attributes. [Speech Technology Magazine]
Kyutai has developed Moshi, an open-source speech-to-speech system that processes overlapping conversation with minimal delay, outperforming GPT-4o and ChatGPT Voice Mode. [DeepLearning.AI]
Microsoft's Copilot Voice has expanded to support 40 languages and enhanced real-time information capabilities. [ZDNet]
SoundHound AI has enhanced its automotive voice assistant with Brand Personalities, allowing automakers to create customized AI personas matching their brand identities. [Speech Technology Magazine]
Document AI
Lender Toolkit has partnered with Ocrolus to integrate advanced document analysis technology into its mortgage underwriting products. [PR Newswire]
Translation
Argos Multilingual, a provider of enterprise localization solutions, announced the general availability of Argos MosAIQ, its advanced AI-powered translation platform. [MultiLingual]
Chinese researchers have tripled their focus on AI speech translation development, according to a new study released in early 2025. [Slator]
Meta has partnered with UNESCO to enhance AI translation capabilities for underserved languages, beginning with Inuit languages through collaboration with Nunavut’s government. [Engadget]
Procentrix has launched TALIS, an AI-powered system that processes multilingual audio, video, and documents for government agencies. [Slator]
Supertext has launched an all-in-one translation platform combining AI-powered free translation with optional expert review, outperforming DeepL in blind testing. [Slator]
Search
Cerebras Systems and Perplexity AI have partnered to launch Sonar, an AI search system running at 1,200 tokens per second on specialized chips. [VentureBeat]
Google's search dominance slipped marginally in 2024, maintaining 95% market share while AI-powered competitors like ChatGPT and Bing made modest gains. [TechRadar]
Google has begun internal testing of ‘AI Mode’, a Gemini 2.0-powered search feature designed to handle complex, exploratory questions with follow-up capabilities. [Search Engine Land]
HigherVisibility has released a study showing that while 71.5% of Americans have tried AI search tools, traditional search engines remain dominant in 2025. [PRWeb]
Perplexity has surpassed Gemini as a preferred AI tool due to its superior search capabilities, accuracy, detailed responses, and better reference handling. [ZDNet]
Trepp has launched new AI tools, including AI search and a chatbot, to enhance how professionals navigate and extract insights from its commercial real estate platform TreppCRE. [PRWeb]
AI in Journalism
Agence France Press has partnered with Kairntech to enhance its multilingual news processing through AI-powered editorial tools for classification and entity extraction. [PRWeb]
Health Tech
AcroDocz has integrated with Clio's legal platform to provide AI-powered medical summaries and billing reports for personal injury law firms. [EIN Presswire]
Aganitha has launched Igniva, a suite of networked AI agents designed to accelerate therapeutic research and drug development through collaborative analysis and reasoning capabilities. [EIN Presswire]
California’s new bill aims to prohibit AI systems from impersonating licensed healthcare providers, addressing concerns after companies like Koko and Character.ai misled users about chatbots’ human status. [Vox]
HealthSci.AI has launched a decentralized science platform enabling researchers to create and customize AI agents for healthcare research, biotech innovation, and scientific collaboration. [EIN Presswire]
Interpreters Unlimited has partnered with Diya Health to launch AI-powered healthcare interpretation and translation services. [Slator]
Life Link III, a provider of critical care air medical transport, and Playback Health, a provider of healthcare technology solutions, announced a strategic partnership that aims to revolutionize clinical documentation within the air medical industry. [Business Wire]
Otto, an all-in-one communication platform for veterinary clinics, has launched its AI Suite, designed to streamline operations, enhance team efficiency, and support collaborative pet care. [Business Wire]
Therapy Brands plans to integrate AI across its product portfolio, starting with TheraNest in 2025, focusing on automated note-taking and treatment planning features. [EIN Presswire]
Legal Tech
Aavenir has launched an interactive tool allowing users to experience Avy, its AI-powered contract management assistant. [EIN Presswire]
CobbleStone Software has partnered with GovSmart to expand its contract management software offerings through government contracting solutions and marketplaces. [PRWeb]
And CobbleStone Software has launched a survey to gather insights from contract management professionals about industry trends and challenges in 2025. [PRWeb]
Stephenson Harwood has partnered with Curvestone to develop a custom generative AI contract review solution for Digital Operational Resilience Act compliance. [Artificial Lawyer]
Elite has launched Validate, an Azure OpenAI-powered billing compliance solution that helps law firms monitor outside counsel guidelines compliance in real time. [Legal IT Insider]
Gavel has expanded its AI-powered document automation tool Blueprint to support 60 languages, up from just English, enabling cross-border legal document generation. [LawSites]
Midpage has achieved full US case law coverage and is preparing to launch its own citator, while expanding its AI-powered legal research platform to target larger law firms. [LawSites]
SimpleDocs has launched SimpleAI, a Microsoft Word-integrated tool enabling legal teams to create and deploy custom AI-powered contract review playbooks for streamlined negotiations. [LawSites]
Funding
Addlly AI has raised SGD1.2M in seed funding to expand its AI marketing content automation platform and accelerate international growth. [FinSMEs]
Affineon Health has raised US$5m from multiple investors to expand its AI-powered healthcare workflow automation platform. [Feed the AI]
Alexi has secured CA$4.5m in debt financing from TD Innovation Partners to enhance its AI litigation platform, following its recent US$15m Series A funding. [FinSMEs]
Breakthrough has secured $600,000 in pre-seed funding to develop its AI-powered B2B sales messaging optimization platform. [Tech Funding News]
ConverzAI has secured US$16m in Series A funding to expand its AI-powered recruitment automation platform internationally and enhance its capabilities. [VentureBeat]
Current AI has launched with US$400m from France, tech companies, and philanthropies to develop AI for public benefit. [Fortune]
Datox AI has secured pre-seed funding to expand its AI-driven regulatory reporting solutions for financial institutions. [FinTech Global]
DevAI has raised US$6m in seed funding to expand its AI-powered network intelligence solutions for enterprise IT teams. [FinSMEs]
Eudia has secured up to US$105m in Series A funding for its Augmented Intelligence platform that helps Fortune 500 legal teams drive strategy and manage risk. [Artificial Lawyer]
Factor has acquired legal technology design firm Theory and Principle to enhance its AI-first legal services. [LawSites]
Fal, a generative media platform for developers, has raised US$49m in Series B funding to expand its development capabilities. [Fortune]
FileAI has raised US$14m in Series A funding to expand its AI-powered file processing platform that handles over 200 million files annually for global enterprises. [FinSMEs]
Groq has secured a US$1.5B commitment from Saudi Arabia to expand its AI chip delivery and data center operations in Dammam. [Yahoo Finance]
Harrison.ai has raised $112M in Series C funding to expand U.S. operations and develop AI-powered medical diagnostic solutions for radiology and pathology. [FinSMEs]
Harvey, a legal AI startup, has secured US$300m in funding while approaching US$100m in annual recurring revenue. [Fortune]
Human.org has raised US$7.3m in pre-seed funding to develop a blockchain-based solution for AI alignment. [FinSMEs]
Keragon has raised US$7.5m in seed funding to expand its AI-powered healthcare automation platform that integrates over 300 software tools for medical professionals. [FinSMEs]
Lingopal.ai has raised US$14m in Series A funding for its real-time speech translation platform that preserves speaker tone across 120+ languages. [FinSMEs]
Model ML has emerged from stealth with US$12m in funding to help professionals make data-driven decisions through its AI platform that automates financial workflows. [Tech.eu]
Noxus secured US$1.5m in pre-seed funding to develop its SaaS platform that helps enterprises build and manage AI workforces. [Tech.eu]
Pinkfish has launched an AI workflow automation platform with US$7.6m in pre-seed funding, helping enterprises build AI agents through natural language prompts. [TechCrunch]
Planted has secured €5m in Seed funding for its AI-powered ESG platform that helps companies automate sustainability reporting and reduce emissions. [Tech.eu]
Positron raised US$23.5m to scale production of its energy-efficient AI inference chips, challenging Nvidia's dominance in the semiconductor market. [Yahoo Finance]
Safe Superintelligence has entered funding talks seeking a US$20B valuation, just months after raising US$1B from major investors. [Yahoo Finance]
SpotDraft has secured US$54m in Series B funding, building on its previous US$26m Series A round, to expand its AI-powered legal tech solutions globally. [Artificial Lawyer]
Thomson Reuters has launched its second C$150m Corporate Venture Capital Fund, focusing on early-stage technology firms in legal, tax, fintech, risk, and media markets. [International Accounting Bulletin]
Tines has raised US$125m in Series C funding to enhance its AI-powered workflow platform, bringing its total funding to US$272m and valuation to US$1.125B. [FinSMEs]
Tofu has raised US$12m in Series A funding to expand its unified AI platform that helps B2B marketing teams create personalized, omni-channel campaigns. [PR Newswire]
There’s More
Anthropic has released a data-driven report showing AI is primarily augmenting rather than replacing jobs, with highest adoption in software development and writing tasks. [VentureBeat]
Epoch AI has determined ChatGPT’s energy consumption is approximately 0.3 watt-hours per query, significantly lower than previous estimates of 3 watt-hours. [TechCrunch]
Google pulled and re-edited its Super Bowl advertisement after Gemini AI falsely claimed Gouda cheese represents 50-60% of global cheese consumption. [TechRadar]
Meta has developed a US$2m brain-signal decoder that translates thoughts into typed text with 80% accuracy, though the half-ton device remains impractical for everyday use. [TechRadar]
Microsoft researchers have found that workers who rely heavily on generative AI tools risk diminishing their critical thinking abilities by shifting focus to verification rather than original analysis. [TechCrunch]
Wired suggests that 2025 is the year of the AI app. [Wired]
The Back Cover
Here’s Joseph Solomon with a cover of John Mayer’s Slow Dancing in a Burning Room.
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!