Sam Altman Responds to Elon Musk's Criticism with New Vision for Open-Source AI

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.

TheGen.AI News

Sam Altman Responds to Elon Musk's Criticism with New Vision for Open-Source AI

OpenAI CEO Sam Altman recently shared his vision for open-source AI in an op-ed for The Washington Post, which aligns closely with what former co-founder Elon Musk has advocated. Altman outlined a four-point plan to enhance U.S. AI leadership, focusing on implementing safeguards, building infrastructure, regulating AI-related exports, and fostering a global community.

A key aspect of Altman's proposal is the importance of open-source AI models in maintaining America's competitive edge. He emphasized that making these models accessible to developers worldwide could strengthen the U.S.'s position in the field. According to Altman, the challenge of leading in AI extends beyond technology to include the values upheld by the technology.

Musk has been critical of Altman's leadership since leaving OpenAI's board in 2018, arguing that OpenAI has strayed from its original mission of being an open-source, nonprofit counterweight to Google. He expressed dissatisfaction with the organization's current direction, which he claims is now focused on maximizing profits and is heavily influenced by Microsoft. Musk even filed a lawsuit against OpenAI in February 2023, accusing the company of violating its nonprofit mission by collaborating with Microsoft, though he later withdrew the suit in June.

Despite his criticism, Musk also recognizes some limitations to openness in AI. In a private email exchange with OpenAI's then-chief scientist Ilya Sutskever in 2016, Musk agreed that there might be a need to be less open as AI development progresses. Sutskever suggested that while sharing everything could be beneficial for recruitment in the short and medium term, not all scientific discoveries should be shared to ensure broader benefits from AI.

Altman's stance on open-source AI might not be an apology to Musk but rather an acknowledgment of the industry's trajectory. Meanwhile, Meta CEO Mark Zuckerberg recently introduced the company's latest AI model, Llama 3.1.

OpenAI Unveils SearchGPT: Entering the AI Search Engine Arena

OpenAI has introduced a new AI search engine prototype, SearchGPT, which aims to compete with Google Search and other search engines. This development comes as no surprise, as AI chatbots have become increasingly integral as search tools. The competition between SearchGPT and other search engines, like Google's AI-enhanced version and Perplexity's Pro Search engine, promises to be fascinating.

The key differences between using SearchGPT and the ChatGPT chatbot for queries are twofold: first, SearchGPT provides the most current information possible, along with relevant links for further reading, and second, it retains context for follow-up queries within the same search session. A demo preview showed that SearchGPT offers visually rich search results, including the source of the information.

OpenAI has not specified if SearchGPT will integrate with existing search engines like Microsoft Bing. The company emphasized the goal of making web searches more efficient by enhancing conversational abilities with real-time information, thereby reducing the effort needed to find relevant results.

Recently, Perplexity upgraded its Pro Search engine to include conversational search and integration with Wolfram|Alpha for academic research. Google has also been layering its Search engine with AI, showcasing features like expanded AI Overviews and multi-step reasoning with Gemini's capabilities at the Google I/O keynote.

Access to SearchGPT is currently limited, with a waitlist for interested users. The search engine is still in the early testing phase, with a select group of users and publishers participating. OpenAI plans to eventually integrate the best features of SearchGPT into ChatGPT.

Stability AI Unveils Stable Video 4D: A New Era in Generative AI

Stability AI is expanding its suite of generative AI models by launching Stable Video 4D, adding a new dimension to its capabilities. While there are numerous AI tools for video generation, such as OpenAI's Sora, Runway, Haiper, and Luma AI, Stable Video 4D offers something unique. Building on the company's existing Stable Video Diffusion model, which transforms images into videos, Stable Video 4D goes further by accepting video inputs and generating multiple novel-view videos from eight different perspectives.

Varun Jampani, the 3D Research team lead at Stability AI, told VentureBeat that they envision Stable Video 4D being used in fields like movie production, gaming, and AR/VR, where dynamically moving 3D objects need to be viewed from various angles.

This isn't Stability AI's first venture into 3D space. In March, they introduced Stable Video 3D, allowing users to create short 3D videos from images or text prompts. Stable Video 4D advances this concept further. While 3D is commonly understood as including width, height, and depth, the 4D aspect adds time, allowing Stable Video 4D to capture moving 3D objects from various camera angles and at different moments.

Jampani explained that Stable Video 4D combines the strengths of the previously released Stable Video Diffusion and Stable Video 3D models, fine-tuning them with a dynamic 3D object dataset. It is a first-of-its-kind model that performs both novel view synthesis and video generation within a single network, unlike existing approaches that use separate networks for these tasks.

Additionally, Stable Video 4D's attention mechanisms are distinct, allowing the generation of each video frame to consider neighboring frames from different camera angles or timestamps, resulting in better 3D coherence and smoothness. Unlike the infill/outfill method used in 2D image generation, which fills gaps in partially completed information, Stable Video 4D synthesizes the novel view videos entirely from scratch, using the original input video as a guide without explicit pixel information transfer.

Apple Commits to AI Safety Guidelines Ahead of Generative AI Platform Launch

Apple has agreed to follow a comprehensive set of voluntary guidelines established by the U.S. to ensure the responsible use of artificial intelligence, as it prepares to launch its own generative AI platform.

According to Bloomberg, Apple joins other major tech companies like Amazon, Google, Meta Platforms (Facebook's parent company), and Microsoft, as well as AI leaders Anthropic, Inflection, and OpenAI. These commitments, initially introduced over a year ago, aim to mitigate the risks associated with AI, particularly generative AI, which has significantly changed user experiences and services.

A statement from July 2023 emphasized that companies developing these technologies must ensure their products are safe, highlighting three core principles: safety, security, and trust. These principles are considered essential for the future of AI and represent a crucial step toward responsible AI development. 

Apple's adherence to these guidelines will play a key role in its future strategies, especially following the introduction of Apple Intelligence at the Worldwide Developers Conference. This marks Apple's formal entry into the generative AI space. The company also announced a long-awaited partnership with OpenAI, integrating the generative AI tool ChatGPT into its Siri digital assistant. ChatGPT will be available for free on iOS 18 and macOS Sequoia, the next versions of Apple's operating systems, later this year.

Apple emphasizes that its generative AI platform prioritizes safety and provides personalized experiences. However, concerns have arisen, including from Elon Musk, who argues that integrating OpenAI at the operating system level poses a security risk. Musk, co-founder of Tesla Motors, has even threatened to ban Apple devices from his companies if the partnership proceeds.

Apple also faces challenges in launching Apple Intelligence in Europe. Although the platform is expected to launch later this year, regulatory uncertainties, particularly concerning the EU's Digital Markets Act, could delay its release. Nonetheless, recent joint statements from EU, UK, and US regulators indicate they are working on "common principles" for AI regulation, which may affect the platform's launch.

TheOpensource.AI News

Chinese AI Companies Turn to Open-Source Models: A Business Strategy Unfolds

In a recent newsletter, the author discussed the accessibility of Chinese large language models (LLMs) and shared their experience testing several of them over the past year. They noted that many people, particularly those unfamiliar with China or the Chinese language, might not know how to access these models. However, the good news is that these Chinese AI models are more accessible overseas than expected. Most can be accessed by registering on their respective websites or using popular open-source AI platforms like Hugging Face. The author has published a practical guide listing a dozen top Chinese LLM chatbots and instructions on how to access them from anywhere in the world.

Through their experimentation, the author observed that while Chinese AI companies generally have stricter access requirements compared to their Western counterparts, a trend toward open-sourcing AI models is making them more accessible globally. For example, Alibaba’s flagship AI foundation model, Qwen (or Tongyi Qianwen), is offered as an open-source model, allowing developers and commercial clients to use it for free. This model, which received a major 2.0 update in June, has gained international recognition and was ranked at the top of Hugging Face’s recent ranking of open-source LLMs, surpassing Meta’s Llama 3 and Microsoft’s Phi-3.

Other Chinese startups, such as DeepSeek and 01.AI, have also made their models open source, achieving high rankings on leaderboards for their LLM products. These companies offer their models for free to users both inside and outside China. The author raises the question of why these companies are opting for open-source AI and how it benefits them as a business decision.

Kevin Xu, a tech investor and founder of Interconnected Capital, explains that Alibaba’s open-source strategy aims to grow its cloud business. If their open-source model becomes popular, more people will use Alibaba Cloud to build AI applications, benefiting the cloud business. Alibaba’s efforts in open-source AI, including releasing models to the public and creating an open-source platform similar to Hugging Face, aim to attract more users to Alibaba Cloud.

For Chinese AI startups not involved in the cloud business, open-source AI offers a proven path to faster commercialization. On the development side, it allows them to adapt established open-source models like Meta’s Llama, accelerating their product development. On the market side, it encourages them to explore alternative model architectures to differentiate themselves from mainstream models.

Eugene Cheah, the San Francisco–based founder of Recursal AI, an open-source AI platform, highlights that the Western approach to improving AI models often focuses on adding more data or scaling them up. This approach is challenging for smaller, newer companies in the LLM industry, as they cannot compete with the resources of giants like OpenAI and Google. The challenge is even greater for Chinese companies due to US export controls limiting access to advanced chips. Cheah notes that Chinese groups, constrained by GPU shortages, are more willing to experiment with unconventional ideas, leading to the development of more efficient and cost-effective models that can attract budget-conscious clients and help these firms find a niche in the AI market.

TheClosedsource.AI News

Coinbase Expands Board with OpenAI Executive and Legal Experts

Coinbase, a U.S. cryptocurrency exchange, has appointed three new members to its board of directors, including Chris Lehane from OpenAI, as part of its efforts to influence U.S. crypto policy. The new board members are Chris Lehane, who is part of the executive team at OpenAI; Paul Clement, a former U.S. Solicitor General under President George W. Bush; and Christa Davies, CFO of Aon and a board member for Stripe and Workday. This expansion increases the board size from seven to ten members.

Coinbase's board expansion comes as the company and the broader cryptocurrency industry seek to become a significant political force in the upcoming presidential election. The industry's future could change if Republican nominee and former President Donald Trump wins the election in November.

Paul Clement will help Coinbase challenge the U.S. Securities and Exchange Commission (SEC) and advocate for clear regulations for digital assets. Chris Lehane, who previously served as a policy chief for Airbnb and was part of the Clinton White House, will provide strategic guidance. Christa Davies will focus on enhancing Coinbase's financial and operational performance globally.

Coinbase emphasized that the new board members bring diverse political perspectives, which is important for bipartisan support of the cryptocurrency industry. Lehane noted the importance of bipartisan support for crypto's success, as reported by Reuters. The Coinbase-backed advocacy group, Stand With Crypto, has gained 1.3 million members.

Additionally, three pro-crypto super political action committees—Fairshake, Defend American Jobs, and Protect Progress—have collectively raised over $230 million to support pro-crypto candidates.

Sam Altman Announces Rollout of Advanced Voice Mode for ChatGPT Plus Users

Earlier this year, OpenAI introduced its latest large language model, GPT-4o, emphasizing its multimodal capabilities and real-time conversational abilities. However, the company stated that these new features wouldn't be released until they passed internal safety and reliability checks. OpenAI mentioned that the enhanced Voice Mode would be available to ChatGPT Plus subscribers sometime in July.

In a recent response on X (formerly Twitter), OpenAI CEO Sam Altman announced that the highly anticipated GPT-4o-powered Voice Mode would enter its alpha phase for paid subscribers next week. The current Voice Mode in ChatGPT is less effective due to delayed responses, with average latencies of 2.8 seconds for GPT-3.5 and 5.4 seconds for GPT-4. In contrast, GPT-4o has no noticeable delay. 

A demo from OpenAI showcased ChatGPT using GPT-4o to teach Portuguese, engage in multi-person conversations, and respond to queries with emotional and non-verbal cues. The rollout of this new Voice Mode was initially planned for a small group of users last month but was postponed to ensure the AI could "detect and refuse certain content."

The new Voice Mode will not be available to all ChatGPT Plus users until the fall. Last week, OpenAI also introduced GPT-4o mini, a streamlined version of the latest large language model, which is more advanced than GPT-3.5 Turbo. Additionally, OpenAI announced a new AI-powered search engine called SearchGPT, which is expected to be integrated with ChatGPT in the future.

OpenAI at Risk of Depleting Cash Reserves Within a Year

OpenAI, the developer of ChatGPT, is projected to face up to $5 billion in losses by 2024, potentially exhausting its cash reserves within a year, according to a report by The Information. The company's expenses on training and inference are expected to reach $7 billion this year, significantly outpacing its revenue. This includes nearly $4 billion spent on Microsoft's servers for ChatGPT inference workloads and up to $3 billion for training new AI models.

Even with substantial discounts from Microsoft Azure, OpenAI's costs have surged due to accelerated development of new AI systems. The company's workforce has expanded to around 1,500 employees, potentially costing $1.5 billion this year—three times the initial projections. This rapid growth has contributed to the company's financial challenges.

OpenAI is generating approximately $2 billion annually from ChatGPT and could earn nearly $1 billion from access fees to large language models (LLMs). However, the total estimated revenue of $3.5 billion to $4.5 billion still leaves a significant shortfall, which may require additional funding within the next year. The Information notes that OpenAI has around 350,000 servers equipped with Nvidia A100 chips for inference, with about 290,000 dedicated to ChatGPT, running at near-full capacity.

Despite these financial challenges, OpenAI CEO Sam Altman remains committed to the company's goal of developing artificial general intelligence (AGI). During a recent talk at Stanford University, Altman expressed indifference to the costs, as long as they find a way to cover them. The company, valued at $80 billion, has raised over $11 billion across seven funding rounds and may need to secure more funding to support its extensive research and development initiatives.

Learn AI in 5 Minutes a Day

AI Tool Report is one of the fastest-growing and most respected newsletters in the world, with over 550,000 readers from companies like OpenAI, Nvidia, Meta, Microsoft, and more.

Our research team spends hundreds of hours a week summarizing the latest news, and finding you the best opportunities to save time and earn more using AI.

Sign up with 1-Click

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!

Keep reading