This Week in NLP #299
What happened in the week ending Friday 9th August 2024.
Above the Fold
Overwhelmed? Here’s our pick for five things to know about this week.
Google’s hiring of Character.ai’s founders highlights Big Tech’s strategy of acquiring top AI talent as the startup AI sector struggles. [Yahoo Finance]
Groq, a semiconductor startup specializing in AI inference chips, raised US$640m in Series D funding to scale its offerings. [Yahoo Finance]
Nvidia’s Blackwell B200 AI chips will face a three-month production delay due to a late design flaw, pushing large shipments to early 2024. [The Verge]
OpenAI co-founder Sam Altman’s social media post featuring strawberries has sparked speculation about a new AI model codenamed ‘Strawberry’, potentially linked to the anticipated GPT-5. [VentureBeat]
Meanwhile, OpenAI faces a crisis with three top executives departing, mounting competitor pressure, and lawsuits, threatening its industry dominance. [VentureBeat]
Now read on for everything else that happened in NLP this week.
Making News:
This Week’s Topics:
The Generative AI Wars
Adept’s investors will recoup their funds after Amazon hired away top employees, sparking regulatory scrutiny over whether it disguised an acquisition. [Semafor]
Apple announced that ChatGPT integration will be available on iOS 18, iPadOS 18, and macOS Sequoia by the end of 2025. [TechRadar]
Apple plans to charge up to US$20 for advanced AI features. [CNBC]
‘You are a helpful mail assistant,’ and other Apple Intelligence instructions. [The Verge]
Character.ai will license its LLM tech to Google for increased funding, with some team members joining Google DeepMind to focus on innovative AI products. [VentureBeat]
Google reduced the price of its Gemini 1.5 Flash model by 80%, making it more competitive than OpenAI’s GPT-4o mini. [Verdict]
Mistral AI announced advancements in model customization, alpha release of Agents, and a stable version of their SDK to improve software development with language models. [Mistral AI]
OpenAI is shifting its DevDay format to low-key, on-the-road sessions focusing on updates to APIs and developer education. [TechCrunch]
ChatGPT’s mobile app achieved its best month ever in July, with US$28m in revenue, driven by the introduction of the faster, multimodal GPT-4o model. [ZDNet]
OpenAI’s GPT-4o update introduces ‘Structured Outputs’ for better data formatting and reduces API pricing to compete with Google’s Gemini models. [ITPro]
Elon Musk filed a new federal lawsuit against OpenAI, alleging it abandoned its mission for profit. [VentureBeat]
The UK Labour government has cancelled £1.3B in tech and AI funding promised by the Conservatives, sparking criticism and concerns about losing competitive edge. [BBC]
Hype Bubble?
Major shifts at OpenAI spark scepticism about impending AGI timelines. [Ars Technica]
The generative AI sector, having soared on initial enthusiasm, faces a reality check with overhyped expectations, expensive computing demands, and the need for practical, incremental progress amidst investor skepticism. [Fortune]
The prolonged AI hype is likely to cause more damage by misallocating resources and creating unrealistic productivity expectations than is currently anticipated by experts. [Fortune]
Hardware
EdgeCortix, with a software-first approach to creating energy-efficient AI processors for edge applications, competes against Nvidia while advancing unique technologies like the Sakura-II AI Accelerator. [VentureBeat]
Daily returns of Humane AI’s US$699 AI Pin are outpacing sales. [The Verge]
Intel passed on investing in OpenAI, missing a critical opportunity that contributed to its struggle in the AI era while rivals Nvidia and AMD surged ahead. [Yahoo Finance]
Kinara’s Ara-2 AI processor successfully runs generative AI applications like Qwen1.5-7B at 12 output tokens per second efficiently on the edge, reducing costs and enhancing PC functionality. [Business Wire]
Samsung’s 8-layer HBM3E chips passed Nvidia’s tests, potentially leading to a supply deal by Q4 2024. [Reuters]
It’s Only a Model
ByteDance launched the AI video app Jimeng AI, joining Chinese tech firms competing with OpenAI’s text-to-video model Sora. [Reuters]
Equall, a Paris-based company, has released new legal-specific LLMs, SaulLM-141B and 54B, claiming they outperform GPT-4 and Mistral AI, and advocating for specialized models in the legal domain. [Artificial Lawyer]
LG AI Research launched South Korea’s first open-source AI model, Exaone 3.0. [VentureBeat]
Why AI researchers are worried about ‘model collapse’. [Bloomberg]
Whose Data?
Leaked documents show Nvidia scraping ‘a human lifetime’ of videos per day to train AI. [404 Media]
Suno admitted to using copyrighted songs to train its AI model but claimed it was legal under fair-use doctrine, while RIAA argued this was industrial-scale infringement. [TechCrunch]
A YouTube creator is suing OpenAI for allegedly using video transcripts without consent to train AI models, seeking over US$5m in damages for affected creators. [TechCrunch]
Ireland’s Data Protection Commission is suing Elon Musk’s X for allegedly using EU user data without consent to train its AI tools, violating GDPR. [Engadget]
The LLM Ecosystem
Efficiencies.AI, an AI as a Service company, has launched EQ AIgent and EfficienciesGPT to help SMEs adapt to AI affordably and efficiently. [AiThority]
Lasso Security has developed Context-Based Access Control to enhance security for Retrieval-Augmented Generation frameworks by contextually evaluating access requests. [VentureBeat]
Qualys has launched TotalAI, a solution designed to securely manage AI and LLM risks by offering comprehensive detection and vulnerability assessment for organizations. [AiThority]
S&P Global is partnering with Accenture to train all 35,000 employees in generative AI, aiming to revolutionize financial data analysis and decision-making. [VentureBeat]
Shaip has launched a generative AI Platform to ensure ethical AI development by providing end-to-end support for data quality, model performance, and regulatory compliance. [EIN Presswire]
SingleStore has launched an AI accelerator in Asia Pacific to help startups develop AI technologies using a real-time data platform, providing credits to eligible startups. [VCWire.tech]
WekaIO, an AI-native data platform company, is working with Contextual AI to provide the data infrastructure underpinning its Contextual Language Models. [MarTech Series]
Other LLM Sightings
Acterys’ latest release, Acterys 24.1, integrates advanced AI with Power BI and Excel to enhance business planning speed, efficiency, and security. [AiThority]
AtScale has integrated its Semantic Layer and Query Engine with LLMs, achieving a 92.5% accuracy in Text-to-SQL translation. [Business Wire]
Dun & Bradstreet introduced SmartMail AI and SmartSearch AI, enhancing D&B Hoovers with generative AI capabilities to improve sales prospecting, targeting, and personalization. [Business Wire]
Eccentex’s new AppBase 8.0 platform leverages generative AI to enable no-code application development. [EIN Presswire]
Ferretly, a social media screening platform, has launched major AI-driven features enhancing screening accuracy, image classifications, results relevancy, flagged post summaries, and detection of rude gestures, weapons, and explicit symbols. [PRWeb]
Qlik announced the availability of Qlik Answers, an AI solution designed to transform unstructured data into actionable business insights through generative AI and advanced analytics integration. [Datanami]
Red Hat’s OpenShift Lightspeed is an AI-powered assistant designed to simplify Kubernetes management by integrating NLP and advanced AI tools directly into the OpenShift console. [ZDNet]
ShareThis’s contextual targeting solution leverages AI technologies to analyze the content of web pages where ads will appear. [MarTech Series]
YouTube is testing a Gemini integration to help select creators brainstorm video ideas, titles, and thumbnails. [TechCrunch]
Risks
Meta’s Prompt-Guard-86M AI safety system can be bypassed by inserting spaces between letters, exploiting its vulnerability to prompt injection attacks. [The Register]
Microsoft’s integration of Copilot AI for productivity in its 365 apps is under scrutiny due to demonstrated vulnerabilities that allow hacker exploitation for malicious activities. [Wired]
OpenAI’s GPT-4o System Card details safety measures and findings from internal and external risk evaluations, deeming the model to be of ‘medium’ risk. [The Verge]
Five US secretaries of state are urging Elon Musk to immediately address misinformation spread by X’s AI chatbot, Grok, regarding 2024 election ballot deadlines. [The Washington Post]
Responses
Anthropic launched an expanded bug bounty program offering up to US$15k for critical AI vulnerabilities, highlighting its focus on AI safety. [VentureBeat]
AppSOC unveiled innovative AI security and governance capabilities to help enterprises manage AI risks and ensure compliance. [Business Wire]
Google has released research that examines generative AI misuse to inform the development of safeguards and promote responsible technology, identifying common exploitation tactics and strategies. [Google]
OpenAI is providing early access to its next AI model to the US AI Safety Institute at NIST, aiming to improve AI evaluation and ensure safety. [The Decoder]
A new technique developed by researchers aims to make it harder to remove safety safeguards from open-source AI models. [Wired]
And Yoshua Bengio has joined the UK-funded Safeguarded AI project to develop AI systems ensuring the safety of critical AI deployments by using AI to monitor AI. [MIT Technology Review]
Regulation
The UK’s Competition and Markets Authority is investigating Amazon’s US$4B investment in AI startup Anthropic for potential antitrust issues. [TechCrunch]
The US Department of Justice has initiated two antitrust investigations into Nvidia concerning its acquisition of Run:ai and potential abuses of market dominance in AI chips. [The Verge]
The US FCC proposes new rules requiring companies to disclose AI-generated robocalls and texts to protect consumers from fraud and election misinformation. [Engadget]
A survey reveals US business leaders increasingly support robust AI regulation due to concerns over data privacy, security, and ethical use of AI technologies. [VentureBeat]
Conversational AI
Figure’s new humanoid robot leverages OpenAI for natural speech conversations. [TechCrunch]
MiaRec has upgraded its conversation intelligence platform using LLM technologies, significantly enhancing topic analysis by recognizing nuanced dialogues accurately without requiring extensive keyword lists. [EIN Presswire]
Here are five things call center AI can do today and what’s on the way. [TechRepublic]
Be Real
Delphi announced its new Video Clone feature, enabling experts and influencers to scale personalized, real-time video interactions globally through advanced digital cloning technology. [Business Wire]
Meta is negotiating deals with high-profile actors and influencers, including Judi Dench and Awkwafina, to integrate their voices into its AI offerings. [The Verge]
OpenAI’s new voice interface for ChatGPT has raised concerns about users forming emotional attachments, among other risks, according to a recent safety analysis. [Wired]
Respeecher and Reality Defender have partnered to enhance audio deepfake detection by integrating Respeecher’s synthetic speech technology with Reality Defender’s cybersecurity expertise. [Speech Technology Magazine]
Voice News
Amazon Music has introduced Topics, an AI-powered feature that allows US users on iOS and Android to browse podcast episodes by specific topics. [Amazon]
Google Lens now supports voice search, allowing users to ask questions or add context to image searches by holding the Lens shutter button. [ZDNet]
OpenAI’s GPT-4o Voice Mode LLM refused to recite fast tongue twisters without pausing, citing the need to breathe like humans. [Futurism]
Phonak has launched the Audéo Sphere Infinio, the world’s first hearing aid with a real-time AI chip designed to improve speech clarity in noisy environments. [Tech Monitor]
Sonde Health has introduced Sonde Cognitive Fitness, a non-invasive solution utilizing vocal biomarkers to assess cognitive effort in real-time through 30-second voice samples. [Speech Technology Magazine]
Document AI
Iron Mountain has launched the InSight Digital Experience Platform, a secure SaaS platform using AI to manage and monetize both physical and digital information. [Business Wire]
Reveille Software and ABBYY have partnered to integrate advanced monitoring and analytics into ABBYY’s Intelligent Document Processing solutions, enhancing efficiency, security, and overall performance. [EIN Presswire]
Zoom has launched Zoom Docs, an AI-powered document collaboration tool, to expand beyond video meetings and compete with Google and Microsoft. [ITPro]
Translation
Language services provider Lionbridge has launched Aurora AI Studio, designed to help companies train data sets to enable advanced AI solutions and applications. [MultiLingual]
Phrase has launched its cloud-based Phrase Localization Platform on AWS Marketplace. [MultiLingual]
Silvia is an AI app that accurately transcribes Spanglish and other mixed languages, addressing a common limitation in current language assistants. [TechCrunch]
Trados has launched its AI Essentials add-on and numerous enhancements to improve translation management, collaboration, and productivity, including a simplified SME-friendly editing mode. [Slator]
Search
Audible’s new AI-powered feature, Maven, allows US users to search for audiobooks using natural language queries and offers AI-generated summaries. [Publishers Weekly]
A federal judge ruled Google has a monopoly on general search services and text ads, violating the Sherman Act, with potential future implications for its business structure. [TechRepublic]
Microsoft is gradually rolling out AI-powered generative search summaries for Bing, providing AI-compiled answers and related information alongside traditional search results. [Engadget]
Reddit will test AI-generated summaries for search results later this year, aiming to help users explore content and find new communities. [TechCrunch]
Writing Assistance
Apple’s iOS 18.1 developer beta introduces AI-driven writing tools with warnings for sensitive content, highlighting Apple’s caution around controversial topics. [TechCrunch]
Automattic has introduced a new AI tool called Write Brief with AI to help WordPress.com bloggers write more clearly and succinctly. [TechCrunch]
OpenAI has developed an AI text watermarking tool but is hesitant to release it due to potential negative impacts on non-English users and benign uses. [TechRadar]
Health Tech
Abridge has partnered with Reid Health to implement AI-driven clinical documentation. [Business Wire]
Amazon’s One Medical team discussed plans to create an LLM called DoctorAI to enhance healthcare efficiency and reduce costs using AI. [Business Insider]
DigitalOwl’s new AI-powered platform streamlines medical record reviews and decision-making for insurance and legal industries. [Business Wire]
MedPeds LLC enhances clinical documentation for in-person and telehealth appointments using eClinicalWorks’ Sunoh.ai AI medical scribe. [Business Wire]
Orchid launched an AI-powered, HIPAA-compliant clinical notes solution that integrates with any EHR to automate note-taking, reducing time spent on administrative tasks for mental health clinicians. [Yahoo Finance]
RingIQ.ai has launched an AI service to answer patient phone calls 24/7 with conversational fluency, improving operational efficiency for healthcare practices. [Business Wire]
Suki AI and Ascension Saint Thomas have partnered to enhance internal medicine residents’ experience by integrating AI technology to reduce documentation time. [Business Wire]
A study found Claude3 more accurate than GPT-4 in medical queries, but both were surpassed by human medical experts. [insideAI News]
Legal Tech
Codex and Flatiron Law Group created a generative AI-driven M&A negotiation simulator to enhance legal training by allowing lawyers to practice negotiations and improve commercial skills. [Artificial Lawyer]
CS Disco Inc announced the general availability of Cecilia Auto Review, a generative AI tool for high-speed, accurate legal document review. [Business Wire]
DeepIP, an AI-based patent assistant integrated with Microsoft Word, enhances patent drafting and response efficiency for law firms and in-house counsels. [Business Wire]
EmotionTrac blends AI with human data for emotion-tracking in legal fields, emphasizing human expertise alongside AI tools like LLMs for accurate jury and trial preparation. [EnterpriseAI]
Gravity Stack, Reed Smith’s legal tech arm, is focusing on generative AI to boost client efficiency and productivity, leveraging tools like OpenAI and Anthropic. [Artificial Lawyer]
HaystackID’s Core Intelligence AI, utilizing generative AI, enhances eDiscovery by automating complex tasks. [AiThority]
Litify launched Litify AI to enhance efficiencies in legal operations through AI-powered document management and insights. [Business Wire]
OpenText has unveiled Axcelerate with Aviator, using generative AI to enhance eDiscovery for legal teams. [PR Newswire]
Pinsent Masons will implement V7 Labs’ generative AI platform ‘V7 Go’ to enhance legal workflows by using LLMs, advanced OCR, and custom automation. [Artificial Lawyer]
Wexler, a generative AI-based legal tech startup, aids in establishing facts in contentious matters. [Artificial Lawyer]
Funding
Bardeen, an AI automation provider, secured US$3m in funding from Dropbox and HubSpot to enhance their AI agents, now used by over 300,000 users. [Business Wire]
BrandRank.AI, a Cincinnati, OH-based company which specializes in generative AI-based brand search analytics, raised US$1.2m in Seed and Angel funding. [FinSMEs]
Contextual AI raised US$80m in Series A funding to advance its model-enhancing AI tool using retrieval augmented generation technology. [Reuters]
Graceview, a Melbourne-based generative AI platform for compliance, raised US$1.5m to expand in Asia-Pacific and Europe. [FinSMEs]
Hedra, backed by US$10m in seed funding, launched its Character-1 AI video creation tool aimed at offering fast, controllable, and expressive long-form video production for content creators and businesses. [Forbes]
Impactpool, an AI-driven career matching platform based in Stockholm, secured US$4m in Series A funding to enhance its mission of connecting talent with impact-driven organizations. [Tech Funding News]
KNIME has raised an additional US$30m to enhance its enterprise-grade AI governance and modelOps capabilities. [insideAI News]
Level AI raised US$39.4m in Series C funding to advance its AI-driven customer experience intelligence and service automation technologies. [FinSMEs]
Napkin, a visual AI platform aimed at simplifying content creation by generating customizable visuals from text, secured US$10m in funding. [TechCrunch]
OpenAI is leading a US$60m Series B round for Opal Camera Inc to fund high-end webcams and develop AI-powered creative devices. [siliconANGLE]
PortfolioPilot.com, an AI-driven financial advisory platform based in San Francisco, raised US$2m in seed funding to enhance its AI capabilities and provide personalized investment insights. [FinSMEs]
ProRata.ai, a generative AI startup that claims it can accurately attribute and share revenues with content owners from AI chatbot subscriptions, has raised a US$25m Series A fundraising round. [Axios]
Polish startup Revoize raised €458k in pre-seed funding to advance its AI-driven speech enhancement technology, aiming to revolutionize remote communication quality. [EU-Startups]
Sweetspot, an AI-powered startup simplifying US government procurement, raised US$2.2m in seed funding to enhance its SaaS tools for navigating government contracts. [Semafor]
Wisedocs, a Toronto-based AI platform for reviewing medical records, received CAD4.5m to expand its client base and product offerings. [FinSMEs]
Acquisitions
Box has acquired Alphamoon’s AI-powered document processing technology to enhance its Intelligent Content Management platform, adding advanced automation and metadata extraction capabilities. [Business Wire]
Hugging Face has acquired XetHub to upgrade its storage backend and enhance its platform’s capacity for hosting larger AI models and datasets with advanced version control. [VentureBeat]
NLX has acquired Radish Systems, integrating its patented ChoiceView technology to enhance multimodal, voice-and-visual customer service interactions for businesses. [NLX AI]
SoundHound is acquiring Amelia AI for US$80m to expand into new verticals and enhance its enterprise services. [TechCrunch]
There’s More
Friend, an AI companion company, spent US$1.8m of its US$2.5m raised on the domain name friend.com. [404 Media]
A proposed AI tool, a ‘digital psychological twin’, aims to aid end-of-life medical decisions by predicting patient preferences based on personal data. [MIT Technology Review]
Google withdrew its controversial Olympics ad featuring its Gemini AI chatbot following backlash for stifling creativity by replacing a child’s words with computer-generated text. [CNN]
Trump suggests avoiding Google in an interview with gamer Adin Ross. [Bloomberg]
The Back Cover
Here's Marie Miller's rendering of Cat Stevens' Peace Train.
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!