This Week in NLP #303
Keep up with what happened in NLP in the week ending Friday 6th September 2024.
Above the Fold
Overwhelmed? Here’s our pick for five things to know about this week.
Amazon's Alexa will incorporate Anthropic's Claude AI, introducing advanced features like complex question answering and personalized responses, but there’ll be a monthly fee. [TechRadar]
Anthropic's launch of Claude Enterprise, a robust AI assistant with a 500,000 token context window and GitHub integration, positions it to challenge major players in the enterprise AI market. [VentureBeat]
OpenAI is contemplating a restructuring to benefit investors. [Yahoo Finance]
OpenAI aims to raise billions, with Nvidia being the more likely investor over Apple. [Yahoo Finance]
Safe Superintelligence, co-founded by ex-OpenAI’s Ilya Sutskever, has raised over US$1B to acquire computing power and hire staff, with a US$5B valuation. [TechCrunch]
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
Anthropic plans to release system prompts for its new Claude feature, Artifacts, after receiving feedback for not including them in the initial release. [VentureBeat]
Apple is developing a tabletop device with a robotic arm featuring a generative AI personality, potentially marking its entry into the robotics and smart home markets. [The Decoder]
Apple's iPhone 16 faces competitive pressures in China, exacerbated by the lack of the Apple Intelligence feature, leaving room for local rivals like Huawei to gain market share. [Yahoo Finance]
TechCrunch takes a look at what Apple Intelligence is, when it’s coming and who will get it. [TechCrunch]
Baidu rebranded its AI app Ernie Bot as Wenxiaoyan to enhance its competitive edge and focus on smarter searches and user value. [Yahoo Finance]
Google’s Gemini AI chatbot for Gmail Q&A is rolling out on Android, with iOS coming soon. [The Verge]
Google is expected to integrate its Gemini AI into Android Auto, which will enhance in-car assistance capabilities. [TechRadar]
And Google's new Chrome update integrates its Gemini AI chatbot into the address bar, potentially reshaping internet interactions. [VentureBeat]
Microsoft's acquisition of Inflection’s team doesn’t raise competition concerns, according to the UK’s antitrust regulator. [TechCrunch]
Microsoft will host a special Copilot event on September 16th to announce the next phase of Copilot innovations, including rebranding and new features for both business and consumer subscriptions. [The Verge]
OpenAI's Japan executive, Tadao Nagasaki, hinted at an upcoming ‘GPT Next’ model, expected to significantly advance LLMs with a release in 2024. [ZDNet]
OpenAI CEO Sam Altman plans to build infrastructure in the US for AI development, costing tens of billions, funded by a global coalition of investors. [Yahoo Finance]
OpenAI's recent improvements to the Assistants API enhance file search controls, helping developers build more autonomous AI agents by allowing fine-tuning of response generation. [VentureBeat]
OpenAI's business offerings reached one million paying users. [VentureBeat]
And ChatGPT now has over 200 million weekly users, with 92% of Fortune 500 companies adopting its tools, despite rising scepticism and organizational prohibitions against generative AI use. [Ars Technica]
Meanwhile, in early August, Meta’s AI assistant had 400 million monthly users and 40 million daily users. [The Information]
The Wall Street Journal observes that the threat to OpenAI is growing. [The Wall Street Journal]
Marc Benioff has announced Salesforce's ‘hard pivot’ to autonomous AI agents with the launch of Agentforce, aiming to revolutionize CRM amidst growing AI competition. [Yahoo Finance]
This piece argues that a global AI arms race isn’t inevitable. [Palladium Magazine]
Hardware
Eurotech is introducing a versatile line of generative AI servers powered by Nvidia AI Enterprise. [AiThority]
Intel's Core Ultra 200V chips, launching September 24th, aim to dominate AI PC processors with enhanced efficiency, AI capabilities, and performance that rivals Qualcomm's Snapdragon and AMD's processors. [Engadget]
Intel and AMD will support Copilot+ AI features, starting in November, enhancing AI PCs powered by their latest chips. [Engadget]
Microsoft will support new AI features on AMD Ryzen AI 300 and Intel Core Ultra 200V laptops through Windows updates, expanding its Copilot+ PC portfolio. [Tom’s Hardware]
Nvidia’s CEO says the new Blackwell chip will have ‘lots and lots’ of supply, following earlier reported snags in production. [Bloomberg]
While Nvidia's Blackwell chip remains unbeatable in AI inferencing performance, competitors like Untether AI and AMD are gaining traction in power efficiency and some benchmarks. [IEEE Spectrum]
And the US Department of Justice is investigating Nvidia for potential antitrust violations related to its AI chips, alleging it penalizes buyers who don’t exclusively use its products. [Engadget]
OpenAI plans to develop its own AI chips using TSMC's 1.6 nm A16 process node to reduce reliance on expensive Nvidia servers. [Yahoo]
Elon Musk has unveiled Cortex, a massive AI supercluster under construction in Tesla’s Texas plant, aimed at advancing real-world AI using 70,000 servers and 50,000 Nvidia H100 GPUs. [TechRadar]
And xAI deployed the world’s most powerful AI training supercomputer, Colossus, with 100,000 Nvidia chips in just 122 days. [Fortune]
Xockets, funded by top tech executives, is suing Nvidia and Microsoft for patent infringement and participating in a buying cartel to lower AI chip prices. [The Verge]
It’s Only a Model
AI2 released OLMoE, their best open-source MoE language model with 1.3 billion active parameters. [Interconnects]
Alibaba Cloud's new Qwen2-VL vision-language model excels in visual data processing, offering multilingual support, video analysis, and open-source variants for commercial use. [VentureBeat]
And Alibaba launched Qwen2-Math and Qwen2-Audio, specialized LLMs for math problem-solving and multi-modal text and audio input, respectively, outperforming state-of-the-art models. [InfoQ]
Cohere has upgraded its Command R series of LLMs to better serve enterprise clients with improved coding, math, reasoning, and data privacy capabilities. [VentureBeat]
Groq’s LLaVA v1.5 7B, a state-of-the-art multimodal AI model supporting image, audio, and text inputs, is now available on GroqCloud. [Groq]
Luma AI's new Dream Machine 1.6 features 12 precise camera motions, improving control and quality in AI video generation. [VentureBeat]
Magic AI's new LTM-2-mini model has a 100 million token context window. [The Decoder]
Meta's Llama models have seen a ten-fold increase in adoption, challenging closed-source AI dominance and pressuring companies like OpenAI to innovate and reduce costs. [VentureBeat]
Microsoft's new open-source Phi-3.5 AI models—Phi-3.5-mini-instruct, Phi-3.5-MoE-instruct, and Phi-3.5-vision-instruct—enhance reasoning, multilingual processing, and visual analysis. [InfoQ]
MiniMax is a new, highly realistic AI video generator from China, known for accurate human movements, backed by Alibaba and Tencent, and currently in competition with Runway Gen-3 and Kling. [Tom’s Guide]
Neurotechnology released its first open-source Lithuanian LLM to advance AI application development and NLP research in the Baltic region. [EIN Presswire]
Nvidia‘s Eagle AI models significantly enhance machines’ visual understanding and interaction capabilities, leveraging high-resolution vision encoding, with wide-ranging applications and open-source availability. [VentureBeat]
And Nvidia will expand AI development services in Japan, focusing on LLMs trained on Japanese data. [MultiLingual]
Stability AI has introduced its top three text-to-image models to Amazon Bedrock, aiming to enhance the platform’s generative AI capabilities and drive enterprise adoption. [The Next Web]
xAI’s Grok-2 language model, released in beta on the X platform and outperforming models like Claude 3.5 Sonnet and GPT-4-Turbo, offers advanced real-time data integration and enhanced features for Premium users, with an API for developers launching later this month. [InfoQ]
Whose Data?
The Dataset Providers Alliance promotes a standardized, ethical approach to AI data licensing, advocating for opt-in consent and transparent compensation structures to support creators. [Wired]
LAION has released Re-LAION-5B, a revised and safer version of their LAION-5B dataset, removing 2236 potentially illegal links in collaboration with multiple safety organizations. [LAION]
OpenAI likely made deals with publishers to avert lawsuits and gain up-to-date data access, positioning itself to potentially disrupt Google's weakening dominance in web search. [The Verge]
X will permanently stop training its AI chatbot Grok on public posts from EU users due to pressure from the region’s data protection regulator. [Engadget]
And researchers have developed the Data Provenance Explorer to address the loss of dataset origins and licensing information. [MIT News]
The LLM Ecosystem
Aleph Alpha's new Pharia AI provides a comprehensive software stack for developing and operating AI applications. [The Decoder]
Couchbase launched Capella Columnar on AWS and Couchbase Mobile with vector search to enhance AI application development by integrating real-time analytics and facilitating edge device searches. [TechTarget]
DataStax announced significant updates to its AI PaaS, enhancing data ingestion, query relevancy, and ease of AI application development through new integrations and features. [Business Wire]
Researchers from Google DeepMind and other institutions have introduced GenRM, a novel approach that leverages LLMs’ generative capabilities to create more effective verifiers for improving accuracy in complex reasoning tasks. [VentureBeat]
Hewlett Packard has introduced HPE Private Cloud AI and new solution accelerators to streamline AI application deployment. [Business Wire]
Impetus Technologies has launched GenAI Innovation Labs to enable rapid development of generative AI solutions for enterprises. [EIN Presswire]
Intuit has significantly enhanced its Generative AI Operating System to accelerate application development, drive innovation, and provide advanced AI-driven experiences across its product suite for approximately 100 million users. [Business Wire]
LM-Kit announced LM-Kit.NET SDK, designed to integrate advanced generative AI into C# and VB.NET applications. [PRWeb]
Neo4j has enhanced its cloud-based graph database, AuraDB, with a generative AI-infused console. [Datanami]
NinjaTech AI announced upgrades to its SuperGPT AI assistant, incorporating multi-modal capabilities for enhanced productivity, including unlimited image generation, coding, writing, and deep research features in a unified interface. [Business Wire]
Progress released MarkLogic Server 12 with enhanced AI response accuracy, simplified secure generative AI app development, and improved data management features. [GlobeNewswire]
Other LLM Sightings
Yi-Coder, a new coding assistant from 01.AI, offers state-of-the-art performance with fewer parameters, challenging larger models. [VentureBeat]
Amazon has launched Rufus, its generative AI shopping assistant, now in beta for select UK mobile app users, offering personalized product recommendations. [Retail Insight Network]
Arhasi's newly released Salesforce Assistant integrates generative AI to provide real-time insights and streamline sales and marketing operations on the Salesforce platform. [EIN Presswire]
AtScale, a provider of semantic layer technology, announced the private preview of its Natural Language Query capabilities, enabling business users to gain immediate data insights by asking questions in plain English. [Business Wire]
Bloomreach has integrated Nvidia NeMo’s generative AI microservices into its ecommerce search and merchandising platform to enhance search relevance and personalization for businesses and consumers. [Business Wire]
Clarivate has launched the Web of Science Research Assistant, a generative AI-powered tool developed to enhance research discovery and analytics using 120 years of data. [AiThority]
HitPaw FotorPea V4.1.1 has introduced an advanced text enhancer, AI painting, and a personalized scenario-based homepage. [PR Newswire]
LinkDR has launched an AI-powered tool to dramatically speed up SEO link building and outreach for marketers. [EIN Presswire]
Sapio Sciences announced enhancements to Sapio ELaiN, an AI-powered lab assistant that uses natural language to streamline lab tasks, enabling scientists to focus on high-value activities. [Business Wire]
Shibumi AI, a new suite of AI-powered tools within the Shibumi platform, enhances the efficiency and effectiveness of strategic initiative management by automating analysis, content creation, prediction, recommendations, and support. [CIO Influence]
Volkswagen is launching an AI voice assistant, IDA, with cloud support from ChatGPT for the 2025 Jetta and others. [The Verge]
WaveCX has launched Curator, an AI-driven search tool enhancing information retrieval and engagement for financial institutions by integrating customized, accurate, and secure search capabilities into their platforms. [Business Wire]
WPAI has launched WP.Chat, an AI assistant for WordPress users offering plugin-specific support, code generation, and product recommendations. [EIN Presswire]
Dating apps are creating AI ‘wingmen’ to help users craft better chat-up lines. [The Financial Times]
Risks and Responses
Google is adding new election-related safeguards to YouTube, Search, Google Play, and AI products in preparation for the upcoming US election, enhancing measures announced in late 2023. [Engadget]
Meta‘s CyberSecEval 3 benchmarks LLMs’ cybersecurity risks, revealing vulnerabilities in Llama 3, and recommends strategies including advanced guardrails, human oversight, and continuous AI security training. [VentureBeat]
Microsoft acknowledges a bug in Windows 11 that incorrectly implies users can remove the upcoming Recall feature, promising a fix in a future update. [The Verge]
Microsoft has urged businesses to manage AI compliance and understand Copilot’s capabilities and limitations before deploying it, highlighting data governance and legal considerations. [TechRadar]
Around half of OpenAI’s AGI/ASI safety researchers, including key leaders, have left due to disagreements over managing superintelligent AI risks and growing influence from the company’s communications and lobbying arms. [The Decoder]
Regulation
The Australian government has proposed 10 mandatory guardrails, including AI testing, human oversight, and the right to challenge automated decisions, to minimize AI risks and build public trust. [TechRepublic]
California’s SB 1047 bill on AI safety, backed by experts and opposed by tech giants, now awaits Governor Newsom’s decision amid significant political and industry pressure. [Vox]
The US, UK, and EU will sign the first legally binding international AI treaty to protect human rights, democracy, and the rule of law. [TechRadar]
Conversational AI
CallRail’s conversational intelligence platform integrates with Typeform to enhance lead attribution and intelligence by combining Typeform submissions with CallRail’s comprehensive tracking, enabling more data-driven marketing decisions. [Business Wire]
Google Photos is enhancing its search function with improved NLP and AI through ‘Ask Photos’, offering a conversational search experience. [Wired]
SoundHound AI has launched new customization tools for its SoundHound Chat AI voice assistant to help automotive brands provide tailored, engaging in-vehicle experiences for their customers. [Business Wire]
Volkswagen has integrated ChatGPT into its cars’ voice assistant, Ida, enhancing its conversational capabilities. [TechRadar]
Be Real
National Novel Writing Month’s acceptance of AI in the writing process has sparked intense debate, drawing significant backlash from prominent authors. [VentureBeat]
YouTube is developing tools to detect face and voice deepfakes, to help artists manage unauthorized AI-generated likenesses. [Engadget]
Voice News
The BBC is trialling AI-generated subtitles for specific shows on its Sounds platform but with human oversight, reflecting cautious adoption of the technology. [TechRadar]
Google has introduced a zero-shot voice transfer module that uses short audio clips to synthesize high-quality voices for individuals with speech impairments. [The Decoder]
Taco Bell plans to expand AI voice ordering to hundreds of US drive-thru locations. [Speech Technology Magazine]
Document AI
Accusoft has integrated an AI-powered Document Q&A feature into its PrizmDoc viewer, enhancing document management efficiency by retrieving precise information from complex documents. [CIO Influence]
Reliant specializes in using AI to automate time-consuming data extraction in research, particularly in literature reviews and scientific studies. [TechCrunch]
Wondershare PDFelement 11 is an AI-powered PDF tool offering comprehensive features for editing, sharing, and collaboration across various platforms. [TechRadar]
Translation
A recent collaborative research project called SignON included the deaf community in the development of a sign language machine translation app. [MultiLingual]
SpeakShift, an AI language translation company, will launch a mobile application, providing instant text, voice, and video translations in over 133 languages. [MultiLingual]
XTM International has released XTM Cloud 13.8, aiming to revolutionize the localization industry with cutting-edge AI-driven tools. [MultiLingual]
Health Tech
Google and others are developing AI that can hear signs of sickness. [Bloomberg]
Netsmart has launched Bells Virtual Scribe, an AI-powered tool using AWS to enhance clinical documentation. [Business Wire]
Paige's AI assistant Alba helps pathologists by consolidating patient data, generating actionable insights, and streamlining report creation. [VentureBeat]
Simplify Healthcare launched Simplify Healthcare AI, a suite of pre-built AI solutions for payers. [Business Wire]
WhizAI’s updated platform integrates a domain-tuned LLM and intent-ready NLP engine to offer life sciences and healthcare companies highly accurate, private, and cost-efficient conversational analytics. [Business Wire]
Legal Tech
Harvey has launched BigLaw Bench, a methodology to evaluate the accuracy of generative AI tools in legal tasks. [Artificial Lawyer]
Morae and ContractPodAi announced a partnership to enhance legal workflows through Leah-powered generative AI, offering advanced contract drafting, review, and analysis capabilities. [EIN Presswire]
PowerPatent's new AI-driven Antecedent Basis Check feature improves patent application quality by ensuring clear antecedent references, reducing rejections and enhancing clarity. [EIN Presswire]
Transpire's integration with Relativity streamlines the workflow from discovery to trial by allowing seamless transfer of data and collaborative analysis. [EIN Presswire]
Ed Tech
David Game College in London is launching the UK’s first AI-taught class, using personalized AI and VR for GCSE students, supported by human ‘learning coaches’. [TechRadar]
Ed Tech companies are promoting AI tools to help teachers with grading and lesson planning. [MIT Technology Review]
Meanwhile, the fear of unreliable AI checkers is leading students to write more robotically to avoid false accusations of cheating. [TechDirt]
Funding
Convin, an AI-powered conversation intelligence platform based in Bengaluru, secured US$6.5m in Series A funding. [The SaaS News]
Cortex, a startup that aims to help software developers be more efficient, has raised US$60m. [Bloomberg]
IdentifAI, a Milan-based startup specializing in AI-generated content detection, raised €2.2m in seed funding to combat deepfakes. [Tech Funding News]
Magic, an AI startup focused on automating software development and code generation, secured US$320m in funding and partnered with Google Cloud to build two AI supercomputers, bringing its total funding to US$465m. [Tech Funding News]
Paradigm, a startup using generative AI, aims to revolutionize spreadsheets by automating data collection and enhancing efficiency, launched with US$2m in seed funding. [VentureBeat]
Cyprus-based startup Placy secured €1m in pre-seed funding to enhance its B2B SaaS tool and AI real estate assistant, aiming for international market expansion. [Tech Funding News]
Seattle startup Revefi raised US$20m in a Series A round to support growth of its software platform that helps companies get a handle on data-related costs, usage and performance. [GeekWire]
Root Signals, a startup with US$2.8m in funding, develops tools to measure, control, and monitor generative AI applications, aiding businesses in reliable AI implementation. [Tech.eu]
Tokyo-based startup Sakana AI raised US$100m in Series A funding to develop nature-inspired AI. [Maginative]
Paris-based AI startup Steerlab has raised US$1.9m in Pre-Seed funding to further develop its platform for automating RFP and security questionnaire responses. [The SaaS News]
You.com, an AI-powered productivity engine, raised US$50m in a Series B funding round to enhance its enterprise AI capabilities. [Business Wire]
Acquisitions
Atlassian acquired Rewatch for its AI meeting notetaker and video tools, planning to integrate it into Loom and Rovo AI to enhance meeting transcripts and actionable insights. [TechCrunch]
Salesforce announced its acquisition of AI voice agent startup Tenyx to enhance its customer service offerings. [Verdict]
There’s More
ABC’s Oprah Winfrey special on AI airing September 12, featuring tech figures like Sam Altman and Bill Gates, has drawn criticism for perceived bias. [Ars Technica]
A study by Cohere demonstrates that including code in LLM pre-training notably enhances performance across various non-coding tasks. [VentureBeat]
Salesforce CEO Marc Benioff criticized Microsoft’s Copilot AI as underperforming and promoted Salesforce’s Agentforce AI as a superior solution for customer service and sales. [Yahoo Finance]
The Back Cover
Here's Lloyiso with a cover of Adele's Easy on Me.
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!