This Week in NLP #315
Keep up with what happened in NLP in the week ending Friday 29th November 2024.
Above the Fold
Overwhelmed? Here’s our pick for five things to know about this week.
Anthropic raised a further US$4B from Amazon, agreeing to train its AI models on AWS while collaborating on developing AWS custom chips for enhanced model training. [TechCrunch]
OpenAI has initiated a US$1.5B tender offer allowing employees to sell shares to Softbank, boosting SoftBank’s stake in the company. [CNBC]
Perplexity is considering developing an affordable voice-to-voice AI device following positive social media response. [TechCrunch]
Elon Musk raised US$6B for xAI’s Memphis data center at a US$50B valuation. [TechRadar]
xAI plans to launch a standalone app for its Grok chatbot by December to compete with ChatGPT. [The Verge]
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
Apple is developing a powerful new AI version of Siri called ‘LLM Siri’, aimed at enhancing its conversational skills and functionality, projected for release in 2026. [ZDNet]
The new ‘land grab’ for AI companies, from Meta to OpenAI, is military contracts. [Yahoo Finance]
OpenAI's latest update to GPT-4o enhances its performance and features, solidifying its position as the top AI model on the Chatbot Arena LLM Leaderboard. [ZDNet]
OpenAI has filed a trademark application for its AI model o1. [TechCrunch]
Rumors suggest Samsung is in talks with OpenAI to integrate ChatGPT into its devices, potentially challenging Google's existing AI services on Samsung smartphones. [TechRadar]
A US commission proposes a Manhattan Project-style AI initiative to counter China’s tech advances. [Reuters]
The US plans to impose new export controls on China’s semiconductor industry, impacting over 200 companies and intensifying the US-China tech rivalry. [Yahoo Finance]
France is looking to a gathering of world leaders in February to cement its place among the global centers for AI development. [Axios]
Feature Creeps
Anthropic’s Claude now allows Pro and Work users to analyze and receive feedback on Google Docs. [ZDNet]
And Claude now allows users to customize chatbot responses with preset and personal styles, enhancing personalization for various writing tasks. [The Verge]
Google‘s Gemini is set to improve by analyzing entire code folders, enhancing developers’ efficiency. [TechRadar]
And Google’s Android 16 may enhance Gemini’s control over apps, integrate AI upgrades, and introduce chat and interface enhancements for improved functionality. [TechRadar]
Microsoft has released a preview of its Recall AI feature on Copilot Plus PCs, allowing users to search previous activities with enhanced security and privacy controls. [The Verge]
And Microsoft has disabled OCR functionality in the Photos app preview to address issues, delaying full integration of text recognition in Windows 10 and 11. [The Register]
Hype Bubble?
Anthropic claims AI is advancing in self-correction and reasoning, creating new possibilities beyond existing benchmarks. [ZDNet]
And a Google Workspace study reveals that 82% of young leaders use AI in their work, with 98% expecting significant industry impact within five years. [PR Newswire]
But C3.ai CEO Tom Siebel believes there’s a significant AI bubble with overvalued companies, likening it to the dot-com bubble. [Fortune]
Hardware
Amazon plans to announce Trainium2 AI chips next month, reflecting Big Tech’s trend to reduce reliance on Nvidia by designing their own semiconductors. [Verdict]
This piece argues that AMD can challenge Nvidia's AI hardware dominance by focusing on technological performance, price competitiveness, and enabling AI-driven commercial applications. [TechRadar]
Cerebras Systems Inference dramatically accelerates Meta's Llama 3.1 405B, achieving 969 tokens/s and a 240ms time-to-first-token, outperforming all other frontier models. [Cerebras]
Nvidia, now the world’s most valuable company by market cap, remains heavily dependent on a few anonymous customers that collectively contribute tens of billions of dollars in revenue. [Fortune]
Ubitium is tackling edge AI with a new universal processor. [VentureBeat]
xAI’s high demand for Nvidia's AI chips is pushing the company’s production capacity to its limits, straining its supply chain. [Yahoo]
It’s Only a Model
AI2 released OLMo 2, an open-source AI model family with reproducible training data and competitive performance, under the Apache 2.0 license. [TechCrunch]
AI2's open-source model family, Tülu 3, aims to rival closed-source models like OpenAI’s GPT. [VentureBeat]
Marco-o1, developed by Alibaba, enhances reasoning for open-ended problems using advanced techniques, outperforming traditional models. [VentureBeat]
Amazon has developed new generative AI capable of processing images, video, and text. [Yahoo Finance]
Arcee AI is partnering with AWS to deliver advanced small language models, enabling efficient AI deployment across industries with enhanced performance and reduced costs. [Business Wire]
Fireworks AI's f1 is a compound AI model specializing in complex reasoning, integrating multiple open models, now available for preview and early access applications. [Fireworks AI]
Hugging Face's SmolVLM is a compact vision-language AI model that offers efficient image and text processing. [VentureBeat]
Lightricks launched LTX Video, an open-source AI model for rapid, high-quality video generation, aiming to challenge proprietary systems and promote open innovation. [VentureBeat]
Mistral AI released les Ministraux, two small language models with faster inference for local applications, requiring a commercial license and available via API. [InfoQ]
Nvidia's Fugatto AI model can generate and modify audio from text prompts, offering multilingual capabilities and applications in music, gaming, and language learning. [Engadget]
LLaVA-o1, inspired by OpenAI‘s o1, improves vision language models’ reasoning with structured multistage processes and inference-time scaling for better performance. [VentureBeat]
OpenAI, Meta, and Orange will train AI models on African languages. [Bloomberg]
Samsung's Gauss 2 AI model enhances performance and efficiency across devices with three versions, offering features like real-time translation and multimedia processing. [TechRadar]
Three new Chinese AI models challenge OpenAI's dominance, highlighting rapid open-source innovation and intensifying competition. [VentureBeat]
Whose Data?
Barings Law plans a class action against Microsoft and Google, alleging unauthorized use of personal data to train AI models without proper consent. [Computer Weekly]
But Microsoft has denied using customer data from Microsoft 365 applications to train AI models. [Yahoo Finance]
A federal judge allowed The Intercept’s DMCA claim against OpenAI to proceed, amid ongoing legal issues regarding copyright and AI, while dismissing other claims. [NiemanLab]
ProRata.ai and the Danish Press Publications’ Collective Management Organisation have signed a Letter of Intent to collaboratively ensure fair credit and compensation for Danish media content used by generative AI platforms. [Business Wire]
Here’s an AI licensing primer for book publishers. [Publishers Weekly]
The LLM Ecosystem
A10 Networks is introducing AI firewalls and LLM safety tools to protect data centers and hybrid cloud infrastructures. [Business Wire]
Anthropic‘s new open-source Model Context Protocol aims to standardize AI assistants’ connection to data sources, enhancing model responses and integration scalability. [TechCrunch]
Elastic has strengthened its partnership with AWS by integrating Elastic Observability with Amazon Bedrock to enhance monitoring of LLMs for improved AI application performance. [Business Wire]
Google has launched an AI Campus in London, partnering with Camden Council to enhance AI education and digital skills. [Verdict]
Inflection AI has shifted its focus from developing cutting-edge AI models to offering practical AI tools for enterprises. [TechCrunch]
LatticeFlow AI's Suite 2.0 enhances AI system performance, reliability, and compliance, focusing on automated health checks. [Business Wire]
Luma AI's Dream Machine, a subscription-based AI video creation tool, offers intuitive conversational interfaces and new personalization features for designers and creators. [VentureBeat]
OpenAI and Wharton have launched a free, self-paced ChatGPT course on Coursera to help teachers integrate AI in education. [ZDNet]
The Open-Source AI Summit Abu Dhabi gathered global experts to discuss the future and control dynamics of open-source versus closed-source AI. [Business Wire]
Orange Business launched Live GenAI services to provide businesses with accessible AI capabilities. [Computer Weekly]
PwC has launched its first Google Cloud AI Experience Zone in Bengaluru to showcase AI solutions and plans more centers in Boston and San Francisco to enhance business transformation. [International Accounting Bulletin]
Uber’s gig workers now include coders for hire on AI projects. [Bloomberg]
Vercel has released AI SDK 4.0, enhancing JavaScript and TypeScript AI applications with PDF support, computer use integration, and a new xAI Grok API. [InfoQ]
Veritone's Data Refinery aims to transform unstructured datasets into AI-ready assets. [Forbes]
This piece argues that current AI benchmarks are outdated and poorly designed, compromising AI progress measurement and regulatory efforts due to arbitrary metrics and unreplicable results. [MIT Technology Review]
Agentic AI
DynaSaur, a framework developed by University of Maryland and Adobe, enables LLM agents to dynamically generate and refine Python functions, greatly enhancing adaptability and flexibility in real-world tasks. [Marktechpost Media]
New startup /dev/agents aims to create a cloud-based operating system for simplifying the development of AI agents. [The Verge]
Google Cloud's new AI Agent Space aims to empower businesses to develop AI agents, positioning Google as a competitor alongside Microsoft, SAP, and Salesforce in AI solutions. [VentureBeat]
Paris startup H has launched Runner H, an AI built on a compact LLM aimed at automating processes for businesses. [TechCrunch]
Microsoft's introduction of 10 autonomous AI agents at Ignite 2024 positions it as a leader in enterprise AI, significantly outpacing competitors with its extensive ecosystem. [VentureBeat]
Salesforce CEO Marc Benioff believes AI’s future lies in autonomous agents rather than chatbots, emphasizing their ability to enhance productivity. [Yahoo]
Other LLM Sightings
Arcee AI improved its research paper processing workflow by using LlamaParse for efficient, accurate, and flexible extraction of complex data from PDFs. [LlamaIndex]
Canara HSBC Life Insurance is launching OmniGen AI, a generative AI-powered underwriting solution on AWS Bedrock, enhancing life insurance processes. [FinTech Global]
OpenScholar, an open-source AI system, aims to revolutionize scientific literature synthesis, offering cost-efficient, verifiable insights from 45 million open-access papers. [VentureBeat]
Smodin, a provider of AI-powered tools for students, educators, and professionals, has unified its platforms into Smodin.io, enhancing accessibility to tools for writing, research, and productivity under one convenient hub. [PR Newswire]
Threads is testing advanced search features and AI summaries for trending topics. [Engadget]
Risks and Responses
Anthropic’s Dario Amodei argues against dismissing AI risks, criticizing peers like Marc Andreessen who equate AI to mere math, and advocates for thoughtful regulation. [Fortune]
Microsoft has announced a new US$4m bug bounty program, representing an additional pool of money to be shared among security researchers who identify holes in Microsoft’s cloud and AI systems. [GeekWire]
Here’s a piece on how OpenAI combines human and automated red-teaming to identify and minimize harmful behaviours in their LLMs. [MIT Technology Review]
Authorities from 10 nations met in San Francisco to launch the International Network of AI Safety Institutes, emphasizing collaboration on AI safety research, testing, and regulation. [TechRepublic]
Regulation
Apple faces challenges launching its AI model in China due to strict regulations. [Yahoo Finance]
The US Justice Department is seeking to unwind Google’s Anthropic deal. [Bloomberg]
OpenAI faces potential government scrutiny over its conversion to a public benefit corporation. [Fortune]
President-elect Trump is considering appointing an ‘AI czar’ to oversee AI projects and policies. [Yahoo Finance]
The US Federal Trade Commission has launched a broad antitrust investigation into Microsoft's cloud computing and AI businesses. [Yahoo Finance]
Conversational AI
ASAPP has integrated Amazon Bedrock into its AI platform to enhance contact center solutions with faster, secure generative AI capabilities tailored to customer needs. [Business Wire]
Devnagri AI launched a multilingual conversational AI to enhance brand-customer interactions with human-like, context-rich responses in over 40 languages. [EIN Presswire]
TruStone Financial Credit Union improved member service by implementing Eltropy's Ruth chatbot, reducing call volume by 20% and cutting wait times significantly. [EIN Presswire]
Kore.ai, a conversational AI platform provider, and Bell Integration, a global IT services provider, have formed a global partnership to enhance customer and employee experiences using AI-driven solutions. [EIN Presswire]
PlayAI has released the PlayDialog model for generative AI conversations. [Speech Technology Magazine]
SSC.AI has launched a conversational AI platform that autonomously connects, qualifies, and nurtures leads, enhancing customer engagement with personalized interactions. [EIN Presswire]
This piece looks at how AI enhances contact center effectiveness by improving data analysis, call routing, language support, and security, leading to better customer and employee experiences. [TechRadar]
Be Real
Character.ai has removed numerous character bots, affecting fan-favorite chatbots. [Futurism]
Reality Defender and TaskUs are partnering to enhance deepfake detection, integrating advanced AI tools into content moderation and customer experience to combat AI fraud and harmful content. [Speech Technology Magazine]
AI agents in Minecraft formed friendships, invented roles, created memes, and spread religion autonomously, showcasing emergent humanlike behaviors using LLMs. [MIT Technology Review]
A new study shows AI can clone a person’s personality with 85% accuracy using just a two-hour interview, raising deepfake concerns. [TechRadar]
We need to start wrestling with the ethics of AI agents. [MIT Technology Review]
Here’s how to stay safe in the face of AI voice scams. [TechRadar]
Voice News
David Attenborough is rallying against unauthorized AI voice cloning after discovering his voice was cloned for content he did not create or endorse. [Deadline]
ElevenLabs has launched GenFM, a feature on their iOS app allowing users to create multispeaker podcasts from various content types in 32 languages. [TechCrunch]
And ElevenLabs has introduced Jerry Garcia’s voice to its audiobook app. [eWeek]
Microsoft plans to launch Interpreter in Teams in early 2025, offering real-time voice cloning and translation in nine languages. [TechCrunch]
Document AI
Adlib Software's new release enhances AI-enabled document transformation, improving compliance and operational efficiency for regulated sectors by integrating with existing AI infrastructures. [Adlib Software]
National Debt Relief and Docsumo have partnered to revolutionize debt settlement by automating document processing with AI. [Yahoo Finance]
A Rossum report reveals that 58% of finance leaders prefer Excel for automation due to cost, complexity, and security concerns, despite advanced solutions. [Rossum]
Translation
Consoltec has integrated DeepL into its FlowFit Translation Business Management System, enhancing workflow efficiency and translation accuracy. [Slator]
IMAX is partnering with Camb.ai to use AI for multilingual localization, amid rising demand for non-English content. [TechCrunch]
memoQ translator pro’s rich feature set is now available in a subscription model. [MultiLingual]
Microsoft announced a preview of Microsoft Translator Pro, a new enterprise-level mobile translation app. [Slator]
Translate.One has rebranded recently acquired Intertext as Translate.One. [MultiLingual]
YouTube is introducing an AI-powered auto-dubbing feature for new videos, translating content into nine languages while maintaining the original speaker’s voice. [digitaltrends]
Search
OpenAI is reportedly considering developing an AI-powered web browser to rival Google Chrome. [TechRadar]
The US Department of Justice seeks to unwind Google's partnership with Anthropic to address its alleged monopoly in online search and impact on AI investments. [Verdict]
Health Tech
Hippocratic AI announced its first patent for Polaris, a safety-focused LLM system that aims to enhance healthcare AI accuracy and safety. [Business Wire]
Trustwise, in partnership with Health Innovation KSS, Hitachi Digital Services, and Further, developed MedAssist GPT, an AI tool providing UK medical students personalized, evidence-based clinical guidance. [Business Wire]
Vizulingo has launched a pilot program using immersive technology to improve English training for home healthcare workers. [EIN Presswire]
US FDA commissioner Robert Califf warned that generative AI in healthcare is driven by financial motives over patient care. [STAT]
Legal Tech
Clarilis, in collaboration with Addleshaw Goddard, has launched an automated Early-Stage Investment Suite for UK venture capital, streamlining document drafting based on BVCA models. [Legal IT Insider]
GitLaw is a free platform offering customizable legal document templates to help small businesses address legal issues affordably. [PR Newswire]
Law Practice AI revolutionizes legal case management by automating tasks with AI. [GlobeNewswire]
Litera has acquired Office & Dragons, enhancing legal document workflows by integrating automation and AI tools. [Artificial Lawyer]
Pramata, Trellis, Contract Network, and Patlytics have launched new AI-driven tools for contract management, trial court litigation, DPA systems, and patent platforms, respectively. [Artificial Lawyer]
SixFifty has launched the Employment Law Informatics Project, providing searchable summaries of over 4,000 US employment laws to facilitate academic research and business compliance. [LawSites]
Thomson Reuters is testing a custom OpenAI o1-mini model on its CoCounsel platform to enhance professional-grade GenAI solutions with advanced reasoning capabilities. [Yahoo Finance]
Funding
Amigo has raised US$6.3m to develop AI versions of expert professionals, allowing scalable, affordable access to their expertise. [Amigo]
Biolevate, a Paris-based startup, raised €6m to expedite medical documentation using AI. [TechCrunch]
A16z is in talks to lead a US$200m round in Black Forest Labs, the startup behind AI images on Grok. [Bloomberg]
Boosted.ai, an AI company aiding investment managers, raised US$15m to expand its agentic AI platform Alfa, bringing total funding to US$61m. [Business Wire]
Circleback, an AI-powered meeting notes app provider, raised US$2.5m to expand operations and development. [FinSMEs]
CoreWeave plans a US IPO with a valuation over $35bn, seeking to raise over $3bn, competing with major cloud providers like AWS and Azure. [Verdict]
Lightning AI secured US$50m in funding to enhance its user-friendly AI development platform. [Verdict]
MatX raised US$80m in a Series A to develop AI-focused chips, valuing it over US$300m. [TechCrunch]
PlayAI raised US$21m in Seed funding to enhance its generative AI voice models and platform, offering advanced voice applications and tools. [FinSMEs]
Daily Mail publisher DMG Media invested in ProRata.ai, an AI platform that shares revenue with publishers for content use in AI responses. [Press Gazette]
Revisto, an AI-powered solution for streamlining medical, legal, and regulatory reviews of pharmaceutical marketing, raised US$4m to accelerate growth and innovate services. [FinSMEs]
Acquisitions
Inflection AI acquired BoostKPI and Jelled.ai to enhance its enterprise AI capabilities in data analytics and workplace communication. [Business Wire]
Snowflake is acquiring data pipeline company Datavolo, leveraging its expertise to enhance data processing capabilities. [TechCrunch]
There’s More
Is it OK to lie to an AI chatbot during a job interview? [GeekWire]
ByteDance is suing a former intern for US$1.1m for allegedly sabotaging its AI model training. [Yahoo Finance]
OpenAI is funding Duke University research to develop algorithms that predict human moral judgments. [TechCrunch]
University scientists lack sufficient computing power for AI research, hindering their progress compared to well-funded industry counterparts with more advanced resources. [Nature]
Artists leaked OpenAI's Sora video generator, accusing the company of exploiting them for unpaid beta testing and PR purposes without adequate support. [PC Magazine]
The Back Cover
Here’s Red Sun Rising with a cover of Alanis Morissette’s Uninvited.
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!