- AI Weekly
- Posts
- AI Weekly: 08/07/23
AI Weekly: 08/07/23
The big players continue to charge ahead...
Good morning and welcome to this week’s edition of AI Weekly! In this week’s news, Google and Microsoft have continued to upgrade their products with the latest AI models, while Dell has also entered the AI landscape by offering hardware and a new platform for organizations to set up access to LLMs.
Meta has also released AudioCraft, an open-source AI music generator that creates audio and music from text prompts.
In cloud-related news, CoreWeave, a GPU-focused cloud compute provider, has secured $2.3 billion in debt financing. On the other hand, Amazon remains incredibly optimistic about its own cloud business, despite AWS growth rates declining over the past three quarters.
Enjoy reading about more of last week’s AI news below!
- ZG
Here are the most important stories of the week:
TEXT
Google has already started exploring a "supercharged" Assistant using technology similar to ChatGPT and Google's own Bard chatbot, beginning with mobile devices. Link.
As part of this change, Google is condensing the Assistant team and eliminating a small number of roles, resulting in layoffs for "dozens" of workers.
The company remains committed to Assistant's bright future and is optimistic about the potential of LLMs to enhance its capabilities.
The specific features that will be introduced to Assistant have not been detailed, but possibilities include answering questions based on web information, similar to the AI chatbot, Bard.
The timeline for bringing this technology to smart home products is not yet clear, and privacy implications are a concern for users.
Microsoft continues to make bold announcements in the AI space, with recent updates on Copilot and Bing Chat capabilities. Link.
Copilot, an AI assistant for Microsoft apps, will be available for $30 per user per month for Microsoft 365 E3, E5, Business Standard, and Business Premium customers.
Bing Chat Enterprise, a work-focused version, allows users to ask work-related questions and get AI-generated answers. It's available in preview form and will later be offered for $5/month per user.
Visual Search via Bing Chat, similar to Google Image Search, is rolling out for desktop and mobile variations.
Microsoft is also introducing Sales Copilot, a variation for sales teams that connects CRM, Microsoft 365, and Teams apps, providing real-time tips and data-driven document generation.
Microsoft's Q4 earnings report showed a 26% growth in generative AI revenue, and the company is investing in AI with the expectation of future payoff. The competitive landscape will likely intensify as other tech giants respond with their own generative AI solutions.
IMAGE/VIDEO
YouTube is testing AI-powered auto-generated video summaries on watch and search pages for a limited number of English-language videos and viewers. Link.
The AI summaries aim to provide a quick overview of videos, helping users decide if a video is relevant to their interests.
YouTube clarifies that the AI summaries do not replace video descriptions, which are still written by creators.
AI-powered video summarizer tools already exist, but some creators find them ineffective for summarizing longer videos.
The impact of AI summaries on YouTube creators is uncertain, and it remains to be seen if the experiment will be widely rolled out.
YouTube continues to explore various AI initiatives, including AI-generated quizzes, AI-powered dubbing, and other AI tools introduced by parent company Google.
Google's AI-powered Search Generative Experience (SGE) is getting a new feature: images and video in the summary box at the top of search results. Link.
SGE is still in the "experiment" phase, but CEO Sundar Pichai sees it as the future of Google Search, allowing them to think outside the traditional search box.
The integration of images and video in SGE aims to synthesize and generate relevant, true, and helpful information for users.
Google is working on making the summary box appear faster and providing more context to the links included.
Surfacing and contextualizing links remain crucial, and Google is displaying publish dates next to articles in the summary box to indicate the recency of information.
Despite the potential of SGE, challenges remain in making it faster and striking the right balance between providing information and letting users find it themselves.
Steg.AI, a company that offers a deep learning-based approach to embed nearly invisible watermarks in digital media, providing proof of ownership and protecting against IP theft, recently raised $5 million in a seed round led by Paladin Capital Group. Link.
The company uses matched machine learning models to customize watermarks to each image, making them imperceptible to humans but easily detectable by their decoding algorithm.
The watermarks are robust and survive transformations like resizing, compression, and deletion of traditional metadata.
Steg.AI's watermarking system works similar to an invisible QR code, allowing it to embed data like URLs, hashes, and plaintext information.
The technology has potential applications for stock photography providers, creators on social media, movie studios, companies handling confidential documents, and more.
Steg.AI is pursuing enterprise-level SaaS integrations to provide seamless watermarking and tracking of media.
Inworld, a startup developing AI to dynamically generate dialogue for games and other applications, has raised more than $50 million from Lightspeed Venture Partners, bringing the total raised to over $100 million at a $500 million post-money valuation. Link.
The AI assistant aims to create lifelike and immersive gaming experiences by allowing developers to link dialog and voice-generating tools to animation and rigging systems in popular game engines.
NPCs powered by Inworld's tech can learn and adapt to new situations, navigate chats with memory and recall, and autonomously initiate goals and actions, adding an element of the unexpected to game experiences.
Inworld has worked with brands like NetEase Games, Niantic, and LG Uplus to build AI-driven NPC experiences and plans to launch an open-source version of its character creation tool, Character Engine, in the coming months.
The startup is eyeing experiences beyond games, such as marketing campaigns, automated customer service agents, and broader entertainment, to tap into a larger addressable market.
SPEECH/AUDIO
Meta released AudioCraft, an open-source AI music generator that creates audio and music from text prompts. Link.
AudioCraft has three models: MusicGen for composing music, AudioGen for sound effects, and EnCodec for advanced audio compression.
MusicGen was trained on Meta-owned and licensed music to address copyright concerns.
Meta aims to compete with OpenAI, Google, and Microsoft by bringing AI-powered tools to the masses, including the open-source Llama 2 language model.
AudioCraft was designed for musicians and sound designers to provide inspiration and facilitate the brainstorming and iteration of compositions.
AudioCraft's approach to text-to-audio generation involves using its proprietary EnCodec neural audio codec to create a new "vocabulary" for the model, making it a more complex undertaking compared to text and image generation.
CODE/DEVTOOLS
SettleMint, a low-code blockchain programming tool, has introduced an AI assistant to its platform to help developers create smart contracts, integrate data, and enhance quality assurance testing. Link.
The AI assistant aims to explain each line of code to developers, helping them understand potential mistakes or vulnerabilities.
The assistant is trained mostly on OpenAI's GPT-4 and is continuously updated with the latest information on blockchain development.
Despite the decline in digital currency prices, SettleMint has seen steady user growth as enterprises recognize the benefits of blockchain technology but lack the technical skills to implement it.
The company's growth in lead generation and signed contracts has multiplied 3-4x this year, reflecting the maturing of the enterprise blockchain space.
SettleMint is expanding into Asia with funding from Fujitsu and witnessing significant growth in the Middle East.
HARDWARE
Dell is entering the generative AI space by offering hardware and a new platform for organizations to set up access to large language models and create generative AI projects. Link.
The company will provide new hardware setups, a managed service platform, and computers to run generative AI projects faster.
Dell partnered with Nvidia for infrastructure, combining Nvidia's Tensor Core GPU with Dell's enterprise AI software and data storage for faster AI model processing.
The company aims to keep up with the growing demand for AI capabilities in businesses and follows other tech companies, including Amazon and Accenture, in offering AI services.
Dell's new computers with built-in AI capabilities are designed to be small and battery-efficient.
Dell's co-COO, Jeff Clarke, sees generative AI as an inflection point driving fundamental change in innovation and customer experience.
Tenstorrent, an AI hardware startup, raised $100 million in a convertible note funding round led by Hyundai Motor Group and Samsung Catalyst Fund. Link.
$50 million of the total funding came from Hyundai's car-making units, Hyundai Motor ($30 million) and Kia ($20 million), to jointly develop chips for future mobility vehicles and robots.
The remaining $50 million was contributed by Samsung Catalyst and other VC funds, bringing Tenstorrent's total raised to $334.5 million.
Tenstorrent focuses on AI processors and licenses AI software solutions and IP around RISC-V.
The company shifted its focus to licensing and services, launching DevCloud, a cloud-based service, and establishing partnerships with India-based server system builder Bodhi Computing and LG.
Tenstorrent faces competition from tech giants like Google, Amazon, Microsoft, and Nvidia in the AI chip race amid a challenging macroeconomic environment and a shortage of components necessary for building AI chips.
HEALTHCARE
AI analysis of mammograms outperformed double reading by two breast radiologists, detecting 20% more cancers without increasing false positives. Link.
The study, conducted with over 80,000 women in Sweden, showed that AI reduced radiologists' workload by 44%.
Breast cancer is the most common cancer in the US, and AI-supported screening has the potential to improve efficacy and reduce interval cancers.
European guidelines recommend double reading for high sensitivity, but the US faces a shortage of breast radiologists.
The study's co-author emphasized the need for further trials and evaluations before implementing AI in mammography screening.
The final results will assess whether AI can reduce interval cancers and justify its use in screening, but these findings will take several years.
AI could help the NHS address a radiologist workforce crisis, speeding up cancer diagnoses and freeing up clinicians' time. Link.
Qure AI has developed platforms to interpret X-rays, CT scans, and ultrasounds, with its qXR platform showing 99.7% accuracy in identifying normal chest X-rays.
The use of AI in radiology could save consultants up to two hours a day by transferring scans to radiographers for reading.
Building trust in AI is crucial for its acceptance in the health service, with education and awareness playing a significant role in dispelling myths and fears.
The UK is leading in adopting AI technology, and continued education and training of doctors are essential for successful AI implementation.
Health Secretary Steve Barclay stated that using AI will improve patient care, and the government is investing in AI imaging tools to support the NHS.
OTHER
Amazon's AWS growth rates have been declining over the past three quarters, reaching 12% year-over-year growth in the most recent quarter, but the company remains optimistic about their cloud business. Link.
Amazon CEO Andy Jassy believes the company may have turned the corner in its cloud business, as more customers focus on driving innovation and bringing new workloads to the cloud.
CFO Brian Olsavsky is optimistic about the future, seeing positive trends continuing into July and anticipating increased revenue from interest in generative AI and large language models.
Jassy highlights that even with the slower growth rate, AWS is still a substantial business, and growing double digits on such a large revenue base requires acquiring many new customers and workloads.
AWS has been benefiting from machine learning and AI business for several years, and the growing popularity of generative AI has the potential to drive even more business in the future.
While short-term guidance for AWS was not provided, it is believed that AWS still has significant room for growth in the long term as more workloads move to the cloud and applications are built on large language models.
CoreWeave, a GPU-focused cloud compute provider, has secured $2.3 billion in debt financing after raising $421 million in recent funding rounds. Link.
The credit facility, led by Blackstone and Magnetar Capital, will provide financial flexibility to meet CoreWeave's expansion goals.
The company plans to use the funds to reach 14 data centers by the end of the year and invest in hardware for contracts with clients.
CoreWeave provides access to over a dozen SKUs of Nvidia GPUs in the cloud, catering to AI, machine learning, visual effects, rendering, and pixel streaming use cases.
The company's infrastructure, hosting as many as ~20,000 GPUs in one location, targets the growing demand for generative AI applications.
Smaller startups, including Inflection AI, are turning to CoreWeave's cloud infrastructure to train their generative AI models, contributing to the company's growth and differentiation.
Apple's approach to generative AI appears cautious, with CEO Tim Cook emphasizing the need for thoughtful development and deployment of such models. Link.
Apple has been relatively quiet on generative AI compared to other tech giants like Microsoft, Google, Meta, and Amazon.
The company has used large language models in limited ways, such as improving autocorrect and dictation in iOS 17, but has avoided using popular AI labels like "GPT" or "large language model."
Apple is reportedly internally testing its own GPT-like engine, similar to ChatGPT and Google's Bard, but is still considering how to incorporate such technology into its products.
New AI technology could pose a risk to Apple's hardware dominance if it makes rivals' devices more attractive.
Apple may have an opportunity to differentiate itself by offering generative AI with a focus on privacy, possibly by running AI processes locally on its devices.
SoftBank has launched SB Intuitions, a new company focused on researching and developing LLMs specialized for the Japanese language. Link.
The company plans to build and sell generative AI services based on Japanese language data.
SB Intuitions will be 100% owned by SoftBank and will use data housed in Japan-based data centers, likely tapping into the company's extensive consumer and enterprise operations in the country.
Hironobu Tamba, a long-time SoftBank employee, will lead the new business, starting with a paid-in capital of 150 million yen (approximately $1 million).
SoftBank will utilize a computing platform built on Nvidia GPUs to develop generative AI and other applications, to be online before the end of March 2024.
SoftBank aims to develop domestically-produced generative AI tailored to Japanese business practice and culture, and plans to establish a multi-generative AI system in collaboration with OpenAI, Microsoft, Google, and other companies.
LA-based Moonware, a company that has developed an AI-powered OS for ground operations in the aviation industry, recently closed a $2.5 million pre-seed financing round led by Third Prime, with participation from Zero Infinity Partners, The House Fund, Lorimer Ventures, Plug and Play, and several angel investors. Link.
Their operating system is called HALO, which algorithmically coordinates ground operations in real-time.
HALO considers variables like distance, departure and arrival times, and crew availability before automatically dispatching crew and equipment to reduce delays, airfield congestion, and plane turnaround time.
The software mines real-time flight information, crew schedules and task allocation, and ground positions and movement of the crew and vehicles.
The startup plans to launch HALO with commercial airlines, starting with airlines that insource their ground operations team, and is in conversation with the U.S. Air Force to address their ground ops coordination challenges.
Moonware aims to introduce autonomy to ground operations, particularly for new forms of transportation like eVTOL and commercial hypersonic flight, later in the decade.
GoStudent, the tutor marketplace, raised $95 million in a strategic fundraise from investors, including Deutsche Bank, Left Lane Capital, DN Capital, Tencent, Prosus, DST, Coatue, and Softbank Vision Fund 2. Link.
The company plans to use the funds to develop enhanced hybrid learning solutions in the DACH region (Germany, Austria, Switzerland), including more online and offline educational support and expanding its virtual reality language learning platform, GoVR.
GoStudent aims to prioritize the creation of AI-driven tools, including an AI lesson plan generator trained on the local curriculum to save tutors time during lessons.
The company currently has around 23,000 tutors on its platform.
CEO Felix Ohswald emphasizes the impact of tailored learning on a child's confidence and the potential to create a bespoke learning path for each student with AI-driven study support and virtual reality lessons.
Left Lane Capital sees potential in GoStudent's vision of hybrid offline and online tutoring, along with synergies from its acquisition of Studienkreis.
Topline Pro, a startup that offers a platform infused with AI to help small home service businesses manage and scale their operations, has raised $12 million in a new Series A funding round led by Forerunner Ventures, with participation from Bonfire Ventures, TMV, and BBG Ventures. Link.
The platform provides tools for businesses to build an online presence, gain trust through reviews and photos, and offer online scheduling and payment options to customers.
The home services category grew 18% in 2021 and 10% in 2022 due to increased demand from homeowners spending more time at home during the pandemic.
Topline Pro raised $5 million in 2022 and has since generated over $180 million in business across thousands of businesses in nearly all 50 states.
The company has expanded its product offerings, including a social media hub and AI-generated Google Ads.
The new capital will be used to expand the team, enhance its generative AI capabilities, and continue product development.