This Week in NLP #302
Keep up with what happened in NLP in the week ending Friday 30th August 2024.
Above the Fold
Overwhelmed? Here’s our pick for five things to know about this week.
OpenAI's Project Strawberry, set to potentially launch this Fall as part of ChatGPT 5, promises enhanced math and programming skills among other advanced capabilities. [TechRadar]
Nvidia, Apple, and Microsoft are in talks to join a funding round for OpenAI that would value the company at over US$100B. [Yahoo Finance]
The California State Assembly and Senate have passed the controversial AI regulation bill, SB 1047, which mandates safety precautions for AI companies before model training. [The Verge]
Nvidia's second-quarter revenue more than doubled to US$30B, largely due to four major customers who collectively contributed 46% of sales. [Fortune]
Is Xi Jinping an AI doomer? [The Economist]
Now read on for everything else that happened in NLP this week.
Making News:
What makes this newsletter different?
This Week’s Topics:
The Generative AI Wars
Amazon’s new subscription service for Alexa AI will launch in mid-October, featuring Smart Briefing, more personalized responses, cooking help, Shopping Scout, and an AI assistant for kids. [TechRadar]
Anthropic has begun publishing system prompts in a bid for transparency and ethical AI. [TechCrunch]
And Anthropic has made its Claude Artifacts feature, which allows users to generate and run interactive code directly in the chat interface, generally available across all tiers. [VentureBeat]
Apple is rumored to be developing a generative AI ‘personality’ for future robotic devices, potentially replacing Siri. [TechRadar]
Noam Shazeer, formerly of Character.ai, has rejoined Google to co-lead its AI project Gemini, integrating Google DeepMind's models into products like Search and Pixel smartphones. [Yahoo Finance]
Google's AI Overviews in India struggle with language accuracy and consistency, leading to incorrect, confusing, and out-of-context search results, especially in Hindi. [TechCrunch]
Google‘s new Prompt Gallery in AI Studio boosts developers’ toolsets with pre-built AI prompts for practical and creative applications. [VentureBeat]
And Google is launching customizable chatbots called Gems for Gemini subscribers, enabling personalized assistants with distinct personalities and roles like gym buddy or cooking partner. [The Verge]
Gmail users on Android can now use the paid Gemini AI assistant to summarize and search emails directly in the Gmail app. [TechCrunch]
Google's Gemini AI will soon integrate into WhatsApp and Google Messages, enhancing message responses and notifications with advanced features. [TechRadar]
Inflection will limit free access to its AI chatbot Pi, enabling conversation exports, while shifting focus towards enterprise products due to resource constraints and post-acquisition changes. [TechCrunch]
OpenAI's significant fundraising efforts suggest it may be rapidly exhausting funds operating ChatGPT and developing its next AI model. [Yahoo Finance]
xAI's Grok-2 and Grok-2-mini LLMs have significantly improved in speed and accuracy following an inference code rewrite, with both models now ranking among the top AI models globally. [VentureBeat]
State-linked Chinese entities are using cloud services from Amazon and rivals to access advanced US chips and AI capabilities, bypassing export restrictions. [Yahoo Finance]
Denmark’s Chamber of Commerce is leading a project to develop a Danish language version of ChatGPT, promoting responsible AI use. [Computer Weekly]
Nations developing AI models in their own languages are driving Nvidia's chip demand, contributing significantly to the company’s revenue. [Yahoo Finance]
Hype Bubble?
AI companies are competing to offer affordable, high-quality models, but face tight margins and the challenge of distinguishing their services beyond raw model sales. [Analytics India Magazine]
AI companies are pivoting from developing advanced models to building practical products but face five key challenges: cost, reliability, privacy, safety, and user interface. [AI Snake Oil]
Hardware
AMD's AI PC strategy focuses on integrating neural processing units in x86 PCs with robust software, performant hardware, and open-source solutions to enable AI-powered features. [Engadget]
Cerebras Systems launched an AI inference service claiming to be the world’s fastest, significantly outperforming competitors like Nvidia, with notable cost efficiency and potential for complex AI tasks. [ZDNet]
CoreWeave and EcoDataCenter are partnering to deploy one of Europe’s first large-scale Nvidia Blackwell clusters in Sweden, enhancing Europe’s AI compute capacity with sustainable solutions. [Tech.eu]
And CoreWeave launched Nvidia H200 Tensor Core GPUs, combining them with advanced infrastructure to accelerate generative AI development and enhance system performance and reliability. [PR Newswire]
The US Commerce Department plans to award Hewlett Packard US$50m to expand and modernize its Oregon facility to boost semiconductor technologies for life sciences and AI applications. [Reuters]
IBM's newly unveiled Telum II processors and Spyre AI accelerators enhance AI capabilities on IBM Z mainframes, addressing escalating AI power demands with improved efficiency, security, and scalability. [EnterpriseAI]
Intel's Gaudi 3 AI accelerator chip will debut with IBM Cloud, targeting hybrid and on-premise environments, amidst stiff competition from Nvidia and AMD. [TechCrunch]
Plaud's new NotePin, a $169 wearable AI audio recorder enabling transcription and summarization, faces challenges like other AI wearables due to accuracy and security issues. [Wired]
SambaNova Systems set a new speed record in AI processing with Meta’s Llama 3.1 405B model, achieving 114 tokens per second. [Analytics India Magazine]
Untether AI's speedAI 240 accelerator cards achieved top performance and energy efficiency in the MLPerf v4.1 benchmarks for both Datacenter and Edge categories. [Business Wire]
It’s Only a Model
Abacus.ai has released Dracarys, an optimized open LLM for coding tasks, significantly improving performance. [VentureBeat]
AI21 Labs’ Jamba 1.5 models excel in long-context tasks, offering superior speed, quality, and multilingual support, and are accessible across multiple platforms under an open model license. [Marktechpost Media]
Aleph Alpha‘s release of open-license LLMs advances EU-compliant, transparent AI development, challenging tech giants’ closed-source approaches. [VentureBeat]
Alibaba Cloud's new Qwen2-VL vision-language model excels in visual data analysis, supports multiple languages, and integrates with third-party tools, boasting strong benchmark performance. [VentureBeat]
APMIC has launched its CaiGunn, its Taiwanese LLM, in the US. [PR Newswire]
Google has introduced improved versions of its Gemini 1.5 AI models, showing substantial performance gains. [VentureBeat]
And Google has relaunched its AI image generation tool with safety features, after pausing it due to inaccurate and offensive depictions of people. [Yahoo Finance]
LG AI Research and Google Cloud have expanded their collaboration to develop and train the EXAONE 3.0 generative AI models and ChatEXAONE using Google Cloud’s AI-optimized infrastructure. [PR Newswire]
Microsoft expanded its Azure AI platform by adding two new multilingual models, Phi-3.5-MoE and Phi-3.5-mini, along with new tools and services. [TechRadar]
Nvidia's Mistral AI-NeMo-Minitron 8B combines pruning and distillation techniques to deliver high accuracy in a compact, efficient model suitable for AI chatbots and virtual assistants. [Nvidia]
Whose Data?
Various prominent news outlets and social platforms are using Apple's Applebot-Extended tool to opt out of contributing their data to Apple’s AI training. [Wired]
Baidu has blocked Google and Bing from scraping its Baike content to protect its data. [Yahoo Finance]
Here’s a summary of ongoing lawsuits against OpenAI and its AI models including claims from various parties alleging copyright infringement, trademark issues, and unauthorized use of content for AI training. [Originality.AI]
TollBit aims to create a marketplace where AI companies pay publishers for data, but its business model is still unproven and speculative. [Engadget]
The LLM Ecosystem
Elastic has integrated its Elasticsearch Open Inference API with Anthropic's Claude models, enabling developers to analyze data in real-time and enhance AI application efficiency. [Business Wire]
Impetus Technologies has launched GenAI Innovation Labs, offering rapid development of enterprise AI solutions through a structured, building block-based approach. [AiThority]
Microsoft has released Prompty, a free Visual Studio Code extension that integrates LLMs like GPT-4o into .NET development workflows. [InfoQ]
MLCommons announced MLPerf Inference v4.1 benchmark results, highlighting new mixture of experts model tests, power consumption insights, and the debut of various AI processors. [Business Wire]
Nous Research's DisTrO optimizer significantly reduces inter-GPU communication, enabling powerful AI training over consumer-grade internet. [VentureBeat]
Nvidia's NIM Agent Blueprints offer developers a free catalog of AI workflows and resources to streamline the creation and deployment of generative AI agents and applications. [VentureBeat]
Pinecone has expanded its serverless vector database offerings to Google Cloud and Microsoft Azure, simplifying deployment for generative AI applications. [Datanami]
Vectara Portal simplifies generative AI development with its no-code, open-source environment, enabling non-developers to create AI applications for various enterprise use cases. [VentureBeat]
Other LLM Sightings
Amazon launched its AI shopping assistant Rufus in India, following its US debut and trailing behind Flipkart’s similar feature by nearly a year. [Yahoo Finance]
Artificial Intelligence Risk and Fynancial have partnered to integrate AIR-GPT technology into Fynancial’s platform, enhancing security, compliance, and efficiency for wealth management advisors. [Business Wire]
Jun Group, a mobile technology company that delivers privacy-first advertising for brands, agencies, and publishers, has unveiled the next evolution of its sentiment analysis tool for influencer marketing campaigns. [MarTech Series]
Google's Gemini AI now offers a note-taking feature for Meet video calls, summarizing key points in a Google Doc available to Workspace customers with specific add-ons. [Engadget]
HubSpot's new AI Search Grader helps marketers optimize their brand’s presence in AI search engines. [Marketing Tech News]
Lumivero's NVivo 15 introduces an AI Assistant to enhance research efficiency, cross-platform compatibility, and data security in qualitative data analysis. [PRWeb]
Semaphore 5.10 enhances productivity by using AI to simplify and speed up semantic knowledge model creation. [GlobeNewswire]
Wyze is testing an AI search feature for Cam Unlimited subscribers that allows keyword searches through camera footage. [The Verge]
Zed, a Rust-based text editor, has introduced AI features powered by Anthropic’s Claude, offering enhanced productivity through an assistant panel and inline transformations. [WebProNews]
Risks
Microsoft Bing Copilot falsely accused German journalist Martin Bernklau of various crimes he reported on, prompting a legal dispute and highlighting AI’s defamation risks. [The Register]
Nearly half of OpenAI's AGI safety researchers have left recently due to concerns that the company is prioritizing product development over safety risks. [Fortune]
Yoshua Bengio worries we may be running out of time to regulate AI. [Bloomberg]
Oklahoma City police are using AI chatbots to write crime reports quickly and accurately, but concerns persist about their reliability in court. [Associated Press]
Responses
Accenture and AWS have jointly launched Accenture Responsible AI Platform to help companies assess and implement responsible AI practices as they plan their generative AI strategies. [VentureBeat]
OpenAI and Anthropic signed an agreement with the US AI Safety Institute to collaborate on AI model safety research, testing, and evaluation. [VentureBeat]
Social platform X modified its AI chatbot Grok to direct users to official election information websites after it was found spreading false data on state ballot deadlines. [Associated Press]
Regulation
Elon Musk supports California’s AI safety bill SB 1047, despite opposition from OpenAI. [TechCrunch]
Ex-OpenAI researchers argue that Sam Altman’s public endorsement of AI regulation is misleading, as the company opposes actual regulatory measures like the SB 1047 bill. [Windows Central]
Meanwhile, OpenAI, Microsoft, and Adobe support California’s AB-3211 bill to label AI-generated content for transparency. [TechRepublic]
And Meta and Spotify criticize EU privacy regulations for hampering AI innovation and call for simplified regulations to enhance open-source AI development in Europe. [TechCrunch]
Some 27% of Fortune 500 companies cite AI regulation as a risk in their annual reports. [The Wall Street Journal]
Conversational AI
Infosys is expanding its collaboration with Nvidia to deploy high-performance generative AI-powered solutions for telcos, enhancing customer experiences and operational efficiency through AI technologies. [AiThority]
Inoria is partnering with Kore.ai to enhance AI-driven self-service and productivity solutions, aiming to meet growing demand for advanced AI-assisted customer interactions. [PRWeb]
Klarna’s CEO plans to significantly reduce its workforce from a peak of 5,000 to 2,000, relying on AI to replace many roles. [Fortune]
Lightspeed Voice is launching NOVA, an AI-powered office virtual assistant platform with features like call transcription, summaries, sentiment analysis, and workflow automation. [PRWeb]
Talkmap’s Talkdiscovery 9.0 conversational AI intelligence service provides visibility into customer conversations to improve data governance, risk and compliance. [MarTech Series]
Here’s a piece on the impact of AI on the Philippines’ call center industry. [Bloomberg]
Be Real
Gannett is shutting down its product reviews site Reviewed amid scrutiny over alleged AI-generated content and staff disputes. [The Verge]
An investigation confirmed AI-generated fake review quotes in the Megalopolis trailer, leading to the removal of marketing team member Eddie Egan. [The Verge]
Journalists in Venezuela, facing severe repression under President Maduro, are using AI avatars to safely report on the country’s deteriorating political situation. [The Guardian]
Voice News
TikTok now allows users to create AI simulations of their own voices for video voice-overs, enabling personalized clips and language translation. [Social Media Today]
The Verge looks at what’s available in the way of transcription apps for keeping detailed records of conversations and meetings. [The Verge]
Document AI
Ascendo AI's new PDF Parser automates the process of scanning and integrating key information from PDF documents. [AiThority]
Enactify.ai has unveiled an AI-powered platform that transforms static documents into interactive, actionable content. [MarTech Series]
IAT Insurance Group has adopted CLARA Analytics’ Claims DocIntel Pro AI platform to enhance efficiency, reduce costs, and prevent errors in their auto and general liability claims processes. [Business Wire]
Proton Scribe, Proton’s privacy-first writing assistant, now writes and proofs emails in eight additional languages. [ZDNet]
Turnitin’s new features provide AI-powered grading, AI writing detection, and personalized feedback tools, aiming to improve educational integrity and foster ethical AI use. [insideAI News]
Translation
BlipCut, an AI-powered video translation tool, has released a new version featuring a subtitle generator that generates and translates subtitles across 95 languages. [PRWeb]
D-ID's AI tool clones your voice and translates your videos into 30 languages with synchronized lip movements for realistic multilingual communication. [TechRadar]
SpeakShift is launching an AI-powered translation app to eliminate language barriers through instant text, voice, and video translations in 133 languages. [EIN Presswire]
Health Tech
Collaborative Drug Discovery has launched the CDD AI Support ChatBot, offering immediate answers to basic queries within its CDD Vault to enhance user support. [EIN Presswire]
Elation Health released Note Assist, an AI-driven feature for its EHR platform that transcribes and structures detailed physician-patient interactions. [Business Wire]
Qventus launched AI Operational Assistants to automate administrative tasks, enhancing healthcare staff productivity and patient care by up to 50%. [AiThority]
SoundHound AI's Amelia Patient Engagement solution, integrated with Epic, powers MUSC Health’s AI agent to enhance patient access and self-service. [Business Wire]
Healthcare contact centers are using AI chatbots for routine tasks and smart routing, and integrating omnichannel platforms with electronic health records for consistent communication. [HealthTech]
Legal Tech
Bloomberg Law has launched an AI-powered enhancement allowing users to view complaint summaries on docket sheets, significantly improving efficiency for legal professionals. [PR Newswire]
Harvey and Icertis have partnered to integrate Harvey’s AI models into Icertis’ CLM platform, enhancing contract management for enterprise users. [Business Wire]
Lexlegis.AI, a Mumbai-based legal research company, launched a legal-specific LLM trained on over 10 million Indian legal documents, aiming to enhance legal research and drafting. [Artificial Lawyer]
Morae Global launched MorAI, a suite of generative AI-powered legal solutions. [EIN Presswire]
Office & Dragons, a legal tech startup, leverages generative AI to automate repetitive legal document tasks. [Artificial Lawyer]
Funding
San Francisco-based Agency secured US$2.6m in pre-seed funding to scale its AI agent observability platform, AgentOps.ai, supporting enterprise-level AI agent deployment and compliance. [PR Newswire]
Bland AI, automating enterprise phone calls with hyper-realistic AI agents, emerged from stealth with US$16m Series A funding to enhance and deploy its platform. [Business Wire]
Creatopy, an AI-driven ad creation platform developed in Romania, has raised US$10m to automate and scale digital ads for over 5,000 clients, focusing on agencies. [TechCrunch]
Cursor raised US$60m in Series A to advance AI-powered coding tools, supporting software development with instant answers, refactors, and bug detection. [Cursor]
Swiss startup FLOWIT, developer of an AI digital coach aimed at non-desk employees, raised €4.2M in seed funding to expand operations and increase its team. [Tech Funding News]
Goodfire raised US$7m to enhance AI model observability using mechanistic interpretability. [VentureBeat]
Inventive AI raised US$4m to enhance its AI-driven RFP tool that optimizes sales workflows and manages enterprise content. [FinSMEs]
Jobilla has secured €6M funding to enhance its AI-driven recruitment tools, automating tasks to significantly reduce hiring campaign time and costs. [Tech.eu]
Katara has raised US$2.2m in seed funding to enhance its AI-agent workflow automation platform targeted at improving developer experience for Web2, Web3, AI, and open-source platforms. [Business Wire]
AI startup Magic raised US$320m to develop advanced code-generating models with Google's and Nvidia's infrastructure. [TechCrunch]
Moveo.AI, a London-based startup specializing in workflow automation and generative AI for enhancing customer experience, raised US$2.6m in seed funding. [Tech.eu]
Otto raised US$6m in seed funding to develop an AI-powered travel assistant designed for business travellers. [Feed the AI]
Reclaim.ai, a startup that uses AI to help companies prioritize tasks and coordinate schedules with a calendar app, has been acquired by file management giant Dropbox. [GeekWire]
Reliant AI launched with US$11.3m to provide AI-driven data analytics for the biopharma industry. [insideAI News]
Rep AI, the developer of an AI Sales Concierge for e-commerce, secured US$8.2m in funding to enhance its tool that improves shopper experience and conversion rates. [Tech Funding News]
Slingshot AI, a mental health startup, raised US$30m in seed funding to enhance its AI-driven behavioral health solutions. [FinSMEs]
Story Protocol raised US$80m in Series B funding to combat AI copyright theft using blockchain, valuing the company at US$2.25 billion. [Tech Funding News]
Kenyan social commerce startup Sukhiba Connect raised a US$1.55m seed extension to expand its WhatsApp-based CRM platform across Africa and other emerging markets. [Techpoint Africa]
Supio, an AI platform provider for personal injury and mass tort law firms, raised US$25m in Series A funding. [FinSMEs]
Viggle AI, a Canadian start-up specializing in AI-driven character animation, raised US$19m in Series A funding to scale and enhance its user-friendly animation tools. [Business Wire]
There’s More
The Open Source Initiative is developing a comprehensive definition for open-source AI. [ZDNet]
Studies show that women are significantly less likely than men to use ChatGPT for productivity, even in the same job roles. [The Economist]
Yale will invest over US$150m in AI development over the next five years to enhance infrastructure, research, education, and interdisciplinary collaboration. [Yale Daily News]
Recent price cuts by Google and OpenAI on text-generating models highlight the commoditization in generative AI, driven by competition, falling inference costs, and large-scale data centers. [TechCrunch]
Google‘s new AI-based system, Ask, for summarizing questions at company-wide meetings is seen by employees as softening inquiries, reducing the meetings’ directness and engagement. [Yahoo Finance]
The Back Cover
Here's India Carney and Magaziine with a cover of Phil Collins' In the Air Tonight.
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!