AI Weekly 07/03/23

Good morning and welcome to this week’s edition of AI Weekly! This past week was filled with numerous funding announcements, headlined by Inflection, Runway, and Typeface’s massive funding rounds.

On the foundation layer, DeepMind announced its large language model, Gemini, to compete with OpenAI’s ChatGPT, while OpenAI faces a lawsuit accusing them of stealing and misusing large amounts of personal data from the internet to train their models.

In our weekly regulation update, the House of Representatives is imposing internal restrictions on the usage of ChatGPT due to privacy concerns, tech leaders across Europe signed an open letter advising policymakers against tight regulation, and the U.S. Department of Commerce is considering additional restrictions on the export of AI chips to China.

Enjoy reading about more on that and other AI news below!

- ZG

Here are the most important stories of the week:

TEXT

Meta has announced plans for behavior analysis systems with AI models "orders of magnitude" larger than existing large language models like ChatGPT’s GPT-4. Link.

• The company provided an overview of the AI models it uses, highlighting its focus on multimodal AI and content recommendation algorithms.

• While Meta has not confirmed actively pursuing models of this size, it aims to efficiently train and deploy very large models at scale.

• The implications suggest that the models are aspirational and likely in development, aiming to understand and analyze user behavior preferences.

• Meta's move aligns with the trend of algorithmic tracking and recommendation seen in platforms like TikTok, aiming to impress advertisers with advanced AI capabilities for understanding user interests and preferences.

• The focus on precise ad targeting reflects the underlying motivation of selling ads and justifying the value and legitimacy of AI-powered targeting systems.

DeepMind, owned by Google, is developing a large language model called Gemini to rival or surpass OpenAI's ChatGPT. Link.

• Gemini aims to combine the strengths of AlphaGo, DeepMind's AI system that defeated a professional human player in the game of Go, with advanced language capabilities.

• Gemini is expected to leverage reinforcement learning techniques to address the limitations of current language models and perform tasks more effectively.

• DeepMind has expertise in reinforcement learning and seeks to apply its knowledge to the field of generative AI.

• Gemini is DeepMind's most ambitious project in the language model space and has involvement from high-level executives at Google.

Inflection AI, an AI startup focused on creating "personal AI for everyone," has raised $1.3 billion in a funding round led by Microsoft, Reid Hoffman, Bill Gates, Eric Schmidt, and new investor Nvidia, valuing the company at $4 billion. Link.

• The funding will support the development of Inflection's first product, an AI-powered assistant called Pi, which provides knowledge based on a person's interests and needs.

• Inflection-1, the AI model powering Pi, is said to be competitive with or superior to models such as OpenAI's GPT-3.5 and Google's PaLM-540B, performing well on various measures but falling behind in coding tasks.

• The company is working with Nvidia and CoreWeave to build a large AI training cluster, leveraging 22,000 Nvidia H100 GPUs.

• With this funding round, Inflection becomes the second-best-funded generative AI startup, behind OpenAI, and surpassing Anthropic.

• Despite the challenging economic environment, generative AI startups continue to attract significant investments, with approximately $1.7 billion generated across 46 deals in Q1 2023.

IMAGE/VIDEO

Zoom is introducing the Intelligent Director feature to cater to hybridized workplaces and ensure better visibility of meeting participants. Link.

• The feature focuses on each participant's face, displays it in a box on the screen, and tracks their movements to avoid the "bowling alley effect" seen in large boardroom meetings.

• The system can track up to 16 meeting members and utilizes machine learning to determine the best camera angle to track participants head-on.

• Zoom clarified that the system does not collect biometric data but simply detects if individuals in front of the camera have the necessary proportions.

• The Intelligent Director feature is not compatible with all webcams and requires specific camera models such as AVer Cam550, Poly e70, and Yealink UVC 86.

• Currently, the feature is supported on Windows and Mac platforms with specific hardware requirements, but Zoom plans to expand availability in the future.

Runway, a startup focused on generative AI tools for multimedia content creators, has raised a $141 million extension to its Series C round. Link.

• The investment comes from notable companies including Google, Nvidia, and Salesforce, among others.

• The additional funding will be used to scale Runway's in-house research efforts and expand its team.

• Runway was founded in 2018 and offers AI-powered tools for moviemakers, cinematographers, and photographers.

• Its flagship product is Gen-2, an AI model that generates videos from text prompts or existing images.

• Runway also has an entertainment division called Runway Studios and organizes the AI Film Festival, showcasing films produced with AI.

Typeface, a generative AI startup focused on enterprise use cases, has raised $100 million in a Series B funding round led by Salesforce Ventures, valuing the company at $1 billion. Link.

• The funding will be used to expand Typeface's platform and grow the company's team to meet the demands of enterprise customers.

• Typeface offers a platform consisting of a content hub, Blend for personalized content generation, and Flow for integration into existing apps and systems.

• The emphasis on brand governance, content safety, and privacy sets Typeface apart from its competitors in the generative AI space.

• While legal challenges may arise regarding copyright protections for AI-generated works, Typeface asserts that its customers own all the assets generated on the platform.

• The enthusiasm around generative AI continues to grow, with a significant demand among enterprises for personalized content generation, leading to increased investment in the space.

SPEECH/AUDIO

Slang.ai is an AI platform developed by Alex Sambvani and Gabriel Duncan to address the issue of overwhelming phone calls for brick-and-mortar businesses like restaurants and retailers. Link.

• Slang acts as a digital phone concierge, automatically answering calls and handling tasks such as reservations and appointment modifications.

• The platform uses automatic speech recognition to understand callers of all ages and different accents, allowing businesses to streamline operations and provide exceptional service.

• Slang provides access to data about why customers are calling, helping businesses identify opportunities and address issues proactively.

• While there are other startups offering AI-powered phone interaction solutions, Slang stands out with its large customer base of over 200 clients and significant revenue growth in 2022.

• Slang recently raised $20 million in funding, which will be used to establish partnerships, expand into new industries, and grow the company's team.

Voice.ai, a startup specializing in real-time voice changing technology, has raised $6 million in its first external funding round. Link.

• The company has more than 480,000 users and offers a library of over 50,000 voice filters.

• Voice.ai's tools are used by gamers, content creators, and Vtubers on various platforms such as TikTok, Zoom, Discord, and more.

• The funding will be used to hire technical talent, develop new SDKs and APIs, expand platform support, and explore new applications like singing.

• Voice.ai aims to democratize access to voice AI technology and make it affordable for consumers.

• The startup differentiates itself by retaining the core elements of a user's speech, including emotion and pacing, while replacing the sound of the voice to create a unique result in real time.

Scriptic, an immersive storytelling platform, has raised $5.7 million in funding to expand its catalog of phone-first interactive games created using generative AI tools. Link.

• The funding will also be used to build a team of TV and gaming talent and introduce user-generated content, allowing external writers to share and monetize their own stories on the platform using AI-enabled creator tools.

• The funding round was led by Bitkraft Ventures, with participation from the Amazon Alexa Fund and Tower 26, among others. Scriptic has raised a total of $8.2 million.

• Scriptic's game "Scriptic: Crime Stories" is known for its interactive phone-first crime drama, where players solve a murder by exploring the victim's phone.

• The London-based studio was founded in 2018 and has been an early adopter of generative AI tools in story production.

• Scriptic aims to become the "Roblox of narrative" and claims to have 1.5 million organic installs of its proprietary app across Android and iOS devices. It releases new content monthly.

Loora, an AI-based language learning startup, has emerged from stealth with $9.25 million in seed funding to improve conversational English skills through voice-based interactions with its iOS assistant. Link.

• Unlike text-based generative AI applications, Loora focuses on audio interactions, allowing users to speak to the AI tutor and receive spoken responses.

• The startup aims to address the gap in mastering conversational English in one-to-one learning environments, bridging the limitations of casual language-learning apps and expensive human tutors.

• Loora utilizes generative AI models to respond to conversation prompts, generate prompts when needed, and provide guidance to improve spoken English.

• The company has already signed up thousands of paying customers and plans to stay focused on English language tuition for now.

• Loora's AI tutor assesses and provides feedback on pronunciation, prosody, fluency, and sentiment, but currently focuses on speaking rather than reading and writing skills.

CODE/DEVTOOLS

Speakeasy, an API infrastructure startup, has emerged from stealth with a $7.6 million seed investment. Link.

• The company aims to simplify the process of creating and distributing APIs, starting with tools to help developers create APIs and assist their users in implementing them.

• One tool, Managed SDKs, uses AI to generate a complete API and necessary documentation from an Open API spec in minutes.

• The second tool helps users implement the API without worrying about the underlying infrastructure, using HashiCorp's Terraform tool.

• Speakeasy plans to build out its platform with additional features and is hiring engineers to support its expansion.

• The round was led by GV, with participation from Quiet Capital, Flex Capital, StoryHouse Ventures, and Firestreak Ventures.

Celestial AI, a startup focused on optical interconnect technology for data transmission in AI and machine learning workloads, has raised $100 million in a Series B funding round. Link.

• The funding round was led by IAG Capital Partners, Koch Disruptive Technologies, and Temasek’s Xora Innovation fund, bringing Celestial AI's total raised to over $165 million.

• Celestial AI's technology enables memory disaggregation, allowing for more efficient utilization of memory resources in data centers and potentially reducing costs.

• The company's photonics-based architecture uses light to transfer data within and between chips, providing higher bandwidth, lower latency, and reduced power consumption compared to optical alternatives.

• Celestial AI's interconnect product is compatible with industry standards and is being engaged with tier-one customers, including hyperscalers and processor and memory companies.

• The company also sells its own AI accelerator chip, Orion, built on Celestial's photonics architecture.

• Celestial AI faces competition in the photonic integrated circuit market, but its differentiated technology and increasing demand for optical connectivity in AI workloads position it for growth.

Faros AI, a platform for software engineers to streamline operational data, has raised $20 million in a Series A funding round led by Lobby Capital, with participation from SignalFire, Salesforce Ventures, and Operator Collective. Link.

• The funding brings Faros' total raised to $36 million and will be used to strengthen its go-to-market strategy, research and development, and customer success initiatives.

• Faros provides a "single-pane" view of a software engineering team's work, goals, and velocity, allowing users to measure an organization's health, tech debt, dependencies, and bottlenecks in processes.

• The platform breaks down time spent on feature development versus tech debt, measures onboarding processes, and tracks seniority and tenure by team and location.

• The funding will support the rollout of Lighthouse AI, Faros' AI engine that uses natural language processing to gather insights about engineering operations from disparate systems.

• Faros anticipates that every aspect of software engineering will be transformed by AI in the next five years and aims to help engineering organizations make that transition with confidence.

BentoML, an AI development framework, has raised $9 million in a seed financing round led by DCM Ventures, with participation from Bow Capital. Link.

• The platform aims to simplify the development, deployment, and scaling of AI applications by abstracting away the complexities of the underlying infrastructure.

• BentoML targets data scientists, DevOps engineers, and developers who work with AI models.

• It enables developers to make AI applications production-ready more efficiently, reducing the development timeline from months to just a few days.

• The platform has gained traction through its open source community, which has grown to over 3,000 members.

• BentoML plans to balance short-term trends with long-term goals to establish its position in the AI market.

MULTIMODAL

Startup Humane, founded by ex-Apple employees Imran Chaudhri and Bethany Bongiorno, has unveiled its first product, the Humane Ai Pin. Link.

• The wearable gadget features a projected display and AI-powered features, offering a range of contextual and ambient computing interactions.

• The Ai Pin aims to provide a screenless and seamless personal mobile computing experience, performing tasks similar to a smartphone with fewer gestures and voice commands required.

• With sensors, a camera, and computer vision-powered software, the Ai Pin can offer email and calendar summaries, language translation, phone calls, and object recognition.

• Humane is collaborating with Qualcomm to develop the Ai Pin's internal hardware, powered by an unnamed Snapdragon chip from Qualcomm's series.

• Humane has also partnered with SK Networks, Microsoft, OpenAI, LG, and Volvo, and has raised $230 million in funding from notable investors.

Josh.ai has launched JoshGPT, a home automation system powered by OpenAI's ChatGPT technology, offering a more advanced and customizable experience than traditional smart home assistants. Link.

• JoshGPT can handle specific questions and nuances that voice assistants like Siri, Alexa, and Google may struggle with, allowing users to add variables and request more detailed information.

• The system offers hands-free access to generative AI, providing a supercharged assistant for home and on-the-go use.

• The Josh.ai system requires professional installation and can range in cost from $4,000 to $14,000. It includes Josh Nano and Micro microphones, a handheld Josh remote, and the Josh app.

• The public launch of JoshGPT introduces new features such as Intelligent Areas, System Setup, and Room Customization, enhancing the customized experience within the automated home.

• Installers play a crucial role in configuring rooms, devices, audio/video settings, and other parameters to ensure personalized and seamless automation for users.

HEALTHCARE

French startup Gleamer has raised €27 million ($29.5 million) in a Series B funding round to further develop its AI software for diagnosing and detecting bone trauma lesions and other conditions in medical scans. Link.

• The company's flagship product, BoneView, serves as an AI companion for bone trauma x-rays, helping radiologists identify anomalies that may be overlooked by the naked eye.

• Gleamer's products have received certifications for use in Europe, and BoneView has obtained FDA clearance in the United States. The company boasts 6,500 users across 650 institutions in 24 countries.

• Gleamer aims to capitalize on the growing demand for medical imaging services, driven by factors such as increasing chronic diseases, under-funded healthcare systems, and aging populations.

• The startup plans to expand its product portfolio to include CT scans and mammography and to grow its team in Europe and the U.S.

• The Series B round was co-led by Supernova Invest and Heal Capital, with participation from several other investors, including XAnge, Elaia, Bpifrance, and Crista Galli Ventures.

Augmedics, a startup that has developed an augmented reality-based navigation platform for spinal surgeries, has raised $82.5 million in a Series D funding round. Link.

• The funding was led by CPMG, with participation from Evidity Health Capital, H.I.G. Capital, Revival Healthcare Capital, and Almeda Ventures.

• Augmedics' platform, called "xvision," combines hardware and software to provide surgeons with augmented reality "x-ray vision" during procedures.

• The company's technology has been approved by the FDA and has already been used in over 4,000 procedures.

• Augmedics plans to use the funding to further develop its platform and expand its commercial presence.

• The company sees opportunities for expansion beyond spinal surgeries, including other musculoskeletal areas and cranial procedures that utilize CT and MRI scanning. However, the focus for now will remain on the current application.

POLICY/LAW/ETHICS

OpenAI is facing a lawsuit accusing them of stealing and misusing large amounts of personal data from the internet to train their AI tools. Link.

• The proposed class action lawsuit claims that OpenAI secretly scraped massive amounts of personal data without notice, consent, or compensation.

• The data scraping allegedly occurred at an unprecedented scale, according to the lawsuit.

• OpenAI and Microsoft, a major investor in OpenAI, have been named as defendants in the lawsuit.

• The lawsuit argues that OpenAI's products use stolen private information, including personally identifiable information, without informed consent or knowledge.

• The lawsuit seeks injunctive relief to halt the commercial use of OpenAI's products and requests financial compensation, referred to as "data dividends," to individuals whose information was used.

The House of Representatives is imposing restrictions on the usage of ChatGPT due to privacy concerns. Link.

• Representatives can only use the ChatGPT Plus version, which has privacy features specifically designed for House data.

• The notice emphasizes that ChatGPT should be used for research and evaluation purposes only, with privacy settings enabled and limited to non-sensitive data.

• Lawmakers are cautious about the risks associated with generative AI and are actively seeking to understand and address them.

• OpenAI CEO Sam Altman appeared before the Senate Judiciary subcommittee on privacy and technology to discuss ChatGPT and the need for rules and safeguards.

• Senators from both parties recognize the transformative potential of AI while emphasizing the importance of understanding its implications and developing appropriate regulations.

Founders, CEOs, VCs, and industry giants across Europe have signed an open letter to the EU Commission, warning that tight regulation could cause Europe to miss out on the generative AI revolution. Link.

• The letter argues that current regulatory proposals in the EU could stifle innovation and threaten the competitiveness of European companies in AI.

• Major businesses including Siemens and Airbus, as well as startups and investors, have signed the letter.

• Critics argue that the proposed laws might heavily regulate foundational AI models regardless of their use cases.

• The signatories express concerns about compliance costs and liability risks, which could lead to companies and investors moving their activities abroad.

• The letter calls for the formation of an EU regulatory body composed of industry experts to monitor the application of new laws and consider technological advances.

The U.S. Department of Commerce is considering additional restrictions on the export of AI chips to China, including from manufacturers like Nvidia, as part of efforts to limit China's progress in AI, particularly in the military sphere. Link.

• These restrictions could potentially prohibit shipments of chips to customers in China and require an export license for certain AI chips, impacting the commercial AI sector in China.

• The export controls implemented in September already limited the sale of Nvidia's advanced A100 and H100 chips to China, prompting Nvidia to develop a less powerful AI chip, the A800, as a workaround.

• Chinese tech firms, anticipating the demand for computational power, have been stockpiling Nvidia's AI chips, with ByteDance reportedly placing orders worth over $1 billion this year.

• The banned A100 chip is being sold on the black market in China for double the regular price.

• The U.S. government is also considering limiting the leasing of cloud services to Chinese AI companies, which could impact their ability to circumvent the chip bans, but it could have unintended consequences for a wider range of Chinese tech companies caught in the crossfire.

CalypsoAI, a startup developing software to test, validate, and monitor AI applications, has raised $23 million in a Series A-1 funding round led by Paladin Capital Group. Link.

• The platform allows businesses to monitor and shape the usage of large language models, such as ChatGPT, and provides insights into model toxicity and user engagement.

• CalypsoAI aims to prevent sensitive company data from being shared on models and identify attacks coming from generative AI tools.

• The startup addresses concerns about the risks and lack of mature AI governance programs in companies.

• There is demand for generative AI guardrail solutions in the corporate sector, as companies seek to prevent accidental exposure of proprietary information and address toxicity and misinformation issues.

• CalypsoAI will use the funding to drive product development, hiring, and go-to-market initiatives, with plans to double its workforce over the next two years.

OTHER

Snowflake and Nvidia have partnered to offer businesses a platform within the Snowflake Data Cloud for creating customized generative AI applications using proprietary data. Link.

• The integration of Nvidia's NeMo platform and GPU-accelerated computing with Snowflake's capabilities allows enterprises to develop large language models for advanced generative AI services like chatbots and search.

• The collaboration enables customers to customize their generative AI models over the cloud while keeping their data secure and reducing costs and latency.

• The partnership aims to create an AI factory that helps enterprises turn their valuable data into custom generative AI models that power innovative applications.

• By utilizing proprietary data and custom LLMs, businesses can maintain a competitive advantage and leverage institutional knowledge for personalized AI models.

• Nvidia's NeMo platform, hosted within the Snowflake Data Cloud, empowers users to build, customize, and deploy generative AI models with billions of parameters, fueling AI-powered applications that deeply understand businesses and their domains.

Amazon's Kindle Unlimited young adult romance bestseller list was inundated with AI-generated books of nonsense. Link.

• The presence of these AI-generated books indicates that people are spamming the platform and finding ways to monetize it.

• Many of the AI-generated books disappeared from the bestseller list after Amazon took action, but they remained available for purchase.

• The incident highlights the potential negative impact on Kindle Unlimited, as the payout for publishing writers may decrease if AI-generated books gain significant reach.

• The screenshots shared by an indie author show that only a few legitimate books were present on the bestseller list, while the rest were clearly AI-generated.

• Although the AI-generated books are no longer on the bestseller list, they are still searchable on the site, allowing users to read samples of the nonsensical content. Amazon claims to investigate and address concerns regarding book content to provide a trustworthy shopping experience.

Microsoft has introduced a new professional certificate on Generative AI as part of its Skills for Jobs program. Link.

• The certificate is awarded to individuals who complete the free Career Essentials in Generative AI program, which includes a course on generative AI.

• The certificate aims to address the increasing need for AI skills in various industries and organizations.

• The program consists of five classes covering the basics of generative AI, AI frameworks, ethical considerations, and more.

• The certificate is currently available in English and will be offered in several other languages in the future, with the courses remaining free through 2025.

Braze, a customer engagement software company, has announced its new generative AI solution called Sage AI by Braze. Link.

• Sage AI enhances automation, personalization, and predictive capabilities for Braze's customers in the mobile app marketing space.

• The AI Recommendation engine within Sage AI uses machine learning to match relevant items from Braze Catalogs with customers who are likely to purchase them, enabling hyper-personalized messages and increased revenue.

• Braze introduces AI Content QA, an AI-driven tool that uses OpenAI's GPT-4 to check messages for tone, structure, grammar, and appropriate language.

• Winning Path, another Sage AI tool, brings optimization and automation into Braze's Canvas Flow journey orchestration tool, enabling marketers to simulate customer journeys and identify the most effective paths.

• Braze plans to introduce additional features such as Personalized Path, A/B Test Prediction, and more, leveraging predictive analytics and generative AI capabilities.

MobileGPT is a startup revolutionizing WhatsApp by integrating OpenAI models GPT3.5-turbo and GPT-4 into chats, offering AI-enhanced document creation, image generation, and more. Link.

• The company aims to disrupt the tech industry by democratizing the power of AI through WhatsApp, making it accessible to millions of users.

• MobileGPT stands out by integrating AI capabilities into a widely used platform and providing familiar interfaces, removing the learning curve.

• The state of the tech industry in 2023 can be defined by integration, as sophisticated technologies like AI are seamlessly woven into everyday applications and platforms.

• MobileGPT chose to participate in HackerNoon's Startup of the Year awards to showcase their user-friendly AI tools and their positive impact on productivity and daily interactions.

Zenarate, an AI simulation training platform designed for customer service agents to improve their skills and performance, has raised $15 million in a funding round led by Volition Capital. Link.

• Agents engage with an AI coach that provides feedback on their tone and soft skills, offering suggestions and scenarios for them to practice and improve.

• The platform requires minimal IT integration, only needing a browser and headset for agents to participate.

• Zenarate faces competition in the sales and agent training software market, but it aims to disrupt traditional training methods with its simulation-based approach.

• The company claims to have over 80 partner companies and "over a million" users as customers.

• Zenarate plans to use the funding to expand its workforce from 92 employees to 110 by the end of the year.