• "Towards AGI"
  • Posts
  • OpenAI Joins Microsoft’s Competitor List in AI and Search Sectors

OpenAI Joins Microsoft’s Competitor List in AI and Search Sectors

Author

Shen Pandi
August 02, 2024

In partnership with

Welcome to Towards AGI, your premier newsletter dedicated to the world of Artificial Intelligence. Our mission is to guide you through the evolving realm of AI with a specific focus on Generative AI. Each issue is designed to enrich your understanding and spark your curiosity about the advancements and challenges shaping the future of AI.

Whether you're deeply embedded in the AI industry or just beginning to explore its vast potential, "Towards AGI" is crafted to provide you with comprehensive insights and discussions on the most pertinent topics. From groundbreaking research to ethical considerations, our newsletter is here to keep you at the forefront of AI innovation. Join our community of AI professionals, hobbyists, and academics as we pursue the ambitious path toward Artificial General Intelligence. Let’s embark on this journey together, exploring the rich landscape of AI through expert analysis, exclusive content, and engaging discussions.

OpenAI Joins Microsoft’s Competitor List in AI and Search Sectors

On Tuesday, Microsoft included OpenAI in its annual report as a competitor, a list that has traditionally featured major companies like Amazon, Apple, Google, and Meta. Despite a long-term partnership where Microsoft serves as OpenAI’s exclusive cloud provider and utilizes its AI models for commercial and consumer products, this new designation shows the two companies are starting to encroach on each other’s territories. Microsoft is also OpenAI’s largest investor, having invested around $13 billion.

In the filing, Microsoft identified OpenAI, the creator of the ChatGPT chatbot, as a rival in AI services, search, and news advertising. OpenAI recently unveiled a prototype search engine called SearchGPT.

Some companies pay OpenAI for model access, while others use Microsoft’s Azure OpenAI Service. Additionally, Microsoft offers its Copilot chatbot through the Bing search engine and Windows operating systems as an alternative to ChatGPT.

An OpenAI spokesperson told CNBC that the relationship remains unchanged and was established with the understanding that competition would occur. The spokesperson confirmed that Microsoft continues to be a good partner.

The year has been eventful. Microsoft CEO Satya Nadella was reportedly not informed before OpenAI’s board ousted CEO Sam Altman in November. Altman was soon reinstated, and OpenAI gave Microsoft a non-voting board seat, which Microsoft relinquished earlier this month.

In March, Nadella appointed Mustafa Suleyman, a DeepMind co-founder and former leader of startup Inflection AI, as CEO of a new unit called Microsoft AI. Several Inflection employees joined Suleyman at Microsoft. Nadella and Altman maintain a close relationship. Nadella told The New York Times, “One of the things I love about Sam is every day he’s calling me and saying, ‘I need more, I need more, I need more.’”

A New Era for AI Development: GitHub Models Empowers Developers with Gen AI

GitHub has long been a key player in the realm of AI for development, but it hasn't always been easy for developers to experiment with new generative AI models. That's about to change with the launch of GitHub Models, a new initiative aimed at simplifying the process for enterprise developers to try out and build applications using generative AI. 

Previously, GitHub made strides with its GitHub Copilot service, offering code completion and suggestion capabilities powered by a single, curated AI model. GitHub Models expands on this by providing direct access to a wider variety of AI models, including Meta’s Llama 3.1, OpenAI’s GPT-4o, Mistral Large 2, AI21’s Jamba-Instruct, Microsoft Phi-3, and models from Cohere. This new service allows developers to experiment with and integrate generative AI models into their applications beyond just code completion.

Mario Rodriguez, GitHub's senior vice president of product, emphasized the growing necessity for applications to incorporate intelligence. He highlighted that GitHub Models aims to reduce the friction developers face when experimenting with and integrating AI models. Previously, developers needed to navigate multiple sites and create various accounts to try out different models. Now, using a GitHub identity, developers can easily explore and access a broad array of generative AI models within the GitHub platform.

Rodriguez stated, "AI is not a fad, it’s here to stay," and emphasized the importance of minimizing friction to foster market growth. GitHub Models provides a centralized catalog of AI models that developers can experiment with, leveraging their existing GitHub identity.

Beyond simplifying experimentation, GitHub Models also facilitates the transition from experimentation to production deployment of AI-powered applications. This process involves using Microsoft’s Azure platform, given GitHub's integration within the Microsoft ecosystem. Developers can start by experimenting with AI models in the GitHub Models playground, then transition to a GitHub Codespace or VS Code developer environment, and finally use an Azure SDK to obtain tokens and API keys necessary for connecting to the Azure platform.

Rodriguez also identified key challenges in enterprise AI deployment: latency, quality of responses, and cost. GitHub Models aims to help developers navigate these challenges by providing an environment for testing and comparison. While industry benchmarks are useful, Rodriguez noted that offline and online evaluations are essential for making the best decisions.

In summary, GitHub Models is set to make it significantly easier for developers to experiment with and deploy generative AI models, addressing key challenges and integrating seamlessly with the GitHub platform and Microsoft Azure for enterprise applications.

NVIDIA's Real-Time Gen AI Powers Rapid 3D Desert World Creation

During the SIGGRAPH Real-Time Live event on Tuesday, NVIDIA researchers showcased the capabilities of NVIDIA Edify, a multimodal architecture for visual generative AI, by creating a detailed 3D desert landscape within minutes. This live demonstration highlighted how generative AI can significantly accelerate the creative process for 3D artists.

In one of the top sessions of the prestigious graphics conference, the researchers demonstrated how an AI agent enabled them to build and modify a desert scene from scratch in under five minutes. This demo illustrated how generative AI could serve as a powerful assistant to artists, streamlining ideation and producing custom secondary assets that would typically need to be sourced manually.

These AI technologies, by reducing the time required for ideation, can greatly enhance the productivity and creativity of 3D artists. For instance, artists can now generate necessary background assets or 360 HDRi environments in minutes rather than hours, allowing for faster concept exploration and workflow acceleration.

Creating a complete 3D scene traditionally involves supporting a primary asset with numerous background elements, finding an appropriate backdrop, and generating an environment map for lighting—tasks that are both complex and time-consuming. Due to time constraints, artists often had to compromise between speed and creative exploration. AI agents can help overcome this by enabling quick concept realization and ongoing iteration to achieve the desired look.

In the Real-Time Live demo, the AI agent instructed an NVIDIA Edify-powered model to produce dozens of 3D assets, including cacti, rocks, and a bull skull, with previews available within seconds. The agent then used other models to create potential backgrounds and layout designs, demonstrating adaptability by swiftly swapping rocks for gold nuggets when creative direction changed.

With the design plan established, the agent rendered the scene as a photorealistic image in NVIDIA Omniverse USD Composer, an app for virtual world-building.

NVIDIA Edify models support creators by focusing on main assets while expediting the creation of background environments and objects. The demo featured two Edify models:

- Edify 3D, which generates editable 3D meshes from text or image prompts, providing previews and rotating animations within seconds to help creators prototype quickly.

- Edify 360 HDRi, which uses prompts to generate high-dynamic range images of nature landscapes for backgrounds and scene lighting.

Additionally, the demo showcased an AI agent powered by a large language model and USD Layout, an AI model generating scene layouts using OpenUSD, a platform for 3D workflows.

NVIDIA also announced at SIGGRAPH that two major creative content companies are enhancing productivity with generative AI tools powered by NVIDIA Edify. Shutterstock has launched its Generative 3D service in commercial beta, enabling rapid prototyping and 3D asset generation from text or image prompts, with its 360 HDRi generator entering early access. Getty Images updated its Generative AI service with the latest NVIDIA Edify version, improving image creation speed, output quality, prompt adherence, and advanced controls.

The 3D objects, environment maps, and layouts generated with Edify models use USD, a standard format for 3D world composition, ensuring compatibility with Omniverse USD Composer. This allows artists to import Edify-powered creations directly into Composer and further refine the scene using popular digital content creation tools.

Real-Time Live is a highly anticipated SIGGRAPH event, featuring real-time applications including generative AI, virtual reality, and live performance capture technology. Watch the event replay below.

D&B Hoovers Gets Smarter with Dun & Bradstreet's Gen AI Capabilities

Dun & Bradstreet, a global leader in business decisioning data and analytics, has announced the launch of SmartMail AI and SmartSearch AI. These new Generative Artificial Intelligence (Gen AI) features, now part of D&B Hoovers, enhance the company's sales intelligence solution, building on the Gen AI innovations introduced over the past year.

“D&B Hoovers SmartMail AI and SmartSearch AI empower sales and marketing teams to act swiftly and unlock their full potential using trusted insights from the industry’s most extensive business and contact data footprint,” said the company.

D&B Hoovers SmartMail AI and SmartSearch AI enhance sales prospecting and lead generation across various channels, improving targeting and personalization to deliver more intelligent customer experiences. They offer a productive and effortless experience for sales and marketing teams. Client feedback has highlighted significant improvements in productivity, process simplification, and speed to market. Currently, 4,000 customers are utilizing these AI-powered capabilities, reflecting Dun & Bradstreet's strong client engagement and commitment to innovation. Both tools are built on AiBE, Dun & Bradstreet’s foundational architecture for rapidly developing, testing, and launching new solutions.

“The new Gen AI features in D&B Hoovers are crucial for delivering intelligent sales and marketing experiences that significantly impact businesses. It's vital to connect with the right buyers at the right time with the right offers,” said Eric Kider, General Manager, Sales & Marketing Solutions at Dun & Bradstreet. “D&B Hoovers SmartMail AI and SmartSearch AI enable sales and marketing teams to act swiftly and unlock their full potential using trusted insights from the industry’s most extensive business and contact data footprint.”

Hoovers SmartMail AI simplifies the outreach process through automated messaging and deployment to highly targeted contacts. It boosts productivity for sales and marketing teams by leveraging AI-optimized messaging for personalized content creation, such as emails to prospects and customers. SmartMail AI tailors messages based on personalized inputs about the contact and their relationship with the sender, adjusts the language in up to 19 global languages, and prepares the message for sending directly from the sender’s email application.

Hoovers SmartSearch AI, an AI-powered chat assistant, helps users quickly build targeted lists of companies and contacts based on specific criteria like country, city, industry, company size, and more. This interactive feature ensures the delivery of the desired audience to maximize reach and response, making it easy to identify and act on targeted opportunities.

Open-Source AI Restrictions: Misguided and Counterproductive

Artificial intelligence policy debates often touch on contentious issues, including the long-standing conflict between open and closed-source systems. This debate has resurfaced with lawmakers in California and Europe attempting to restrict "open-weights AI models."

Open-weights models, akin to open source software, are publicly accessible systems allowing their underlying code to be inspected and modified by various parties for diverse purposes. Some critics argue that open-sourcing algorithmic models or systems is "uniquely dangerous" and should be restricted. However, imposing arbitrary regulatory limitations on open-source AI systems could have significant drawbacks, such as stifling innovation, competition, and transparency.

This issue gained renewed attention following significant announcements from both government and industry. On July 30, the Commerce Department released a major report on these models, mandated by the AI executive order signed by President Joe Biden in October.

The report generally supports open-weight AI systems and "outlines a cautious yet optimistic path" for them. It concludes that "there is not sufficient evidence on the marginal risks of dual-use foundation models with widely available model weights to conclude that restrictions on model weights are currently appropriate, nor that restrictions will never be appropriate in the future."

In related news, the Federal Trade Commission issued a statement on open-weights models, noting that they "have the potential to drive innovation, reduce costs, increase consumer choice, and generally benefit the public."

These positive statements from the Biden administration are echoed by J.D. Vance, the Republican candidate for vice president, who supports open-source AI as a way to counter Big Tech. This indicates bipartisan support for open-source AI.

White House Report Advocates for Open-Source AI

The White House acknowledges the importance of open source for AI development, a sentiment shared by many businesses utilizing this technology.

On Tuesday, the National Telecommunications and Information Administration (NTIA) released a report advocating for open-source and open models to foster innovation in AI, while also stressing the importance of vigilant risk monitoring.

The report suggests that the US should continue to support AI openness, develop new capabilities for monitoring potential AI risks, and avoid restricting the availability of open model weights.

Additionally, the Swiss federal government mandates that its software be released as open source.

The NTIA report highlights several key benefits of open AI models:

  1. Broader Accessibility: Open-weight models enable developers to build on and adapt existing work, making AI tools more accessible to small businesses, researchers, nonprofits, and individuals.

  2. Innovation Promotion: Openness in AI systems fosters competition and innovation. The report outlines a roadmap for responsible AI innovation and American leadership by embracing openness.

  3. Accelerated Development: Open models can hasten the dissemination of AI benefits and the pace of AI safety research.

  4. Democratization of AI: Open models expand access to powerful AI tools across various sectors and user groups.

  5. Transparency and Understanding: Open models enhance the understanding of AI systems, crucial for effective and reliable development.

  6. Economic Benefits: The widespread availability of US-developed open foundation models can promote innovation and competitiveness, serving the national interest.

  7. Research Advancement: Open models facilitate academic research on AI internals, enabling deeper study and improvement of the technology.

  8. Local Deployment: Open weights allow users and organizations to run models locally on their edge devices, benefiting specific applications and use cases.

  9. Customization: Open models enable creative modifications to meet specific user needs and applications.

These conclusions were based on feedback from government employees, industry leaders, and individuals responding to a Request for Comment on AI model issues. For instance, the Electronic Privacy Information Center (EPIC) recommended balancing the advantages, disadvantages, and regulatory challenges of AI models across the openness spectrum.

GitHub also supports this approach, advocating for the use of open source and open weights in AI, while considering both potential harms and benefits.

Google DeepMind Releases Powerful and Compact Gemma 2 2B AI Model

Google DeepMind unveiled a new, compact, open-source AI model, Gemma 2 2B, on Thursday. Despite its relatively small size of 2.6 billion parameters, this language model outperforms larger counterparts such as OpenAI’s GPT-3.5 and Mistral AI’s Mixtral 8x7B.

Designed to be compact, Gemma 2 2B can fit on a wider range of devices, including smartphones, while still delivering performance comparable to GPT-3.5.

Independent AI research organization LMSYS, which tested Gemma 2 2B, reported that it scored 1130 in evaluation, slightly surpassing both GPT-3.5 Turbo-0613 and Mistral-8x7B, despite these models having over ten times the parameters.

According to Google’s Developer Blog, Gemma 2 2B was created using distillation techniques, which reduce computational demands by transferring knowledge from larger models to smaller ones.

In June, Google DeepMind announced the larger Gemma 2 9B and 27B models but has since shifted focus to developing smaller, more efficient models, anticipating growth in the mobile and edge-based AI market.

Developers can access Gemma 2 2B on the Hugging Face platform, with implementation available via PyTorch and TensorFlow.

aiOla Introduces Rapid ‘Multi-Head’ Model, Surpassing OpenAI Whisper

Israeli AI startup aiOla has introduced a new open-source speech recognition model, Whisper-Medusa, which is 50% faster than OpenAI’s well-known Whisper. Whisper-Medusa builds on Whisper, employing a novel “multi-head attention” architecture that allows for predicting far more tokens simultaneously compared to OpenAI’s version. Its code and weights are available on Hugging Face under an MIT license, permitting both research and commercial use.

Gill Hetz, aiOla’s VP of research, told VentureBeat, “By releasing our solution as open source, we encourage further innovation and collaboration within the community, potentially leading to even greater speed improvements and refinements as developers and researchers contribute to and build upon our work.”

This development could lead to AI systems capable of understanding and responding to user queries in near real-time.

Despite the rise of foundation models that can generate diverse content, advanced speech recognition remains crucial. This technology drives key functions across sectors like healthcare and fintech—assisting with tasks like transcription—and powers highly capable multimodal AI systems. Last year, OpenAI led the field with its Whisper model, which converts user audio into text, processes the query with a large language model, and then converts the response back into speech.

Whisper has set the standard in speech recognition, handling complex speech with various languages and accents in near real-time. It has seen over 5 million downloads each month and supports tens of thousands of applications.

However, aiOla claims its new Whisper-Medusa model can recognize and transcribe speech even faster than Whisper, enhancing speech-to-text conversion efficiency.

To create Whisper-Medusa, aiOla modified Whisper’s architecture by incorporating a multi-head attention mechanism. This technique allows the model to simultaneously attend to information from different representation subspaces at various positions using multiple “attention heads” in parallel. This change enables the model to predict ten tokens per pass instead of one, boosting speech prediction speed and generation runtime by 50%.

Crucially, despite its increased speed, Whisper-Medusa maintains the same level of accuracy as the original Whisper. Hetz noted that they are the first in the industry to apply this approach to an ASR model and make it publicly available for further research and development.

“Improving the speed and latency of LLMs is much easier than with automatic speech recognition systems. The encoder and decoder architectures present unique challenges due to the complexity of processing continuous audio signals and handling noise or accents. We addressed these challenges with our novel multi-head attention approach, resulting in a model with nearly double the prediction speed while maintaining Whisper’s high accuracy,” he said.

Exploring the Future: Generative AI Event

The second event in the Towards AGI series, Exploring the Future: Generative AI became a huge success which was held on July 31. The round-table discussion, which delved into productionized GenAI use cases spanning Legal AI, the Public Sector, the Gen AI stack from AWS, and the journey Towards AGI

The host thoroughly enjoyed the insights and conversations from Chan Nyein Zaw, Mark Longhurst, and Sabrina Pervez, representing Robin AI, Tenacium DC, and Amazon Web Services (AWS). They look forward to many more such discussions in the future.

A huge thanks was extended to Freshminds for hosting the event and providing excellent hospitality. Special thanks were given to James Callander, Julia Gosling, Marija Globarevic, and Henrietta Bamford.

FREE AI & ChatGPT Masterclass to automate 50% of your workflow

More than 300 Million people use AI across the globe, but just the top 1% know the right ones for the right use-cases.

Join this free masterclass on AI tools that will teach you the 25 most useful AI tools on the internet – that too for $0 (they have 100 free seats only!)

This masterclass will teach you how to:

  • Build business strategies & solve problems like a pro

  • Write content for emails, socials & more in minutes

  • Build AI assistants & custom bots in minutes

  • Research 10x faster, do more in less time & make your life easier

You’ll wish you knew about this FREE AI masterclass sooner 😉

In our quest to explore the dynamic and rapidly evolving field of Artificial Intelligence, this newsletter is your go-to source for the latest developments, breakthroughs, and discussions on Generative AI. Each edition brings you the most compelling news and insights from the forefront of Generative AI (GenAI), featuring cutting-edge research, transformative technologies, and the pioneering work of industry leaders.

Highlights from GenAI, OpenAI, and ClosedAI: Dive into the latest projects and innovations from the leading organizations behind some of the most advanced AI models in open-source, closed-sourced AI.

Stay Informed and Engaged: Whether you're a researcher, developer, entrepreneur, or enthusiast, "Towards AGI" aims to keep you informed and inspired. From technical deep-dives to ethical debates, our newsletter addresses the multifaceted aspects of AI development and its implications on society and industry.

Join us on this exciting journey as we navigate the complex landscape of artificial intelligence, moving steadily towards the realization of AGI. Stay tuned for exclusive interviews, expert opinions, and much more!