OpenAI Reports Iranian Group’s Use of ChatGPT to Interfere in US Election

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.

Exploring The Idea Of World’s 1st GenAI-Powered Pen with Auto-Pilot Capability and Multi-Lingual Support

With just 5 sentences, this pen learns your handwriting, captures your tone, and then writes on its own – in any language! Would you be interested in this innovative tech?

TheGen.AI News

OpenAI Reports Iranian Group’s Use of ChatGPT to Interfere in US Election

OpenAI announced on Friday that it had shut down accounts linked to an Iranian group that was using its ChatGPT chatbot to create content aimed at influencing the U.S. presidential election and other issues. The operation, named Storm-2035, leveraged ChatGPT to produce content on topics like commentary on U.S. election candidates, the Gaza conflict, and Israel’s involvement in the Olympic Games, which was then shared through social media accounts and websites, according to OpenAI.

The investigation by the Microsoft-backed AI company revealed that ChatGPT was being used to generate both long-form articles and shorter social media posts. However, OpenAI stated that the operation had minimal impact, as most social media posts attracted little to no engagement, and there was no significant sharing of web articles.

The accounts involved have been banned from OpenAI’s services, and the company is continuing to monitor for any further attempts to breach its policies. 

Earlier in August, a Microsoft threat intelligence report identified Storm-2035 as an Iranian network running four websites disguised as news outlets, actively targeting U.S. voter groups on both sides of the political spectrum with polarizing messages on issues such as the U.S. presidential candidates, LGBTQ rights, and the Israel-Hamas conflict.

As the November 5 presidential election approaches, Democratic candidate Kamala Harris and Republican candidate Donald Trump remain in a tight race.

In May, OpenAI also disclosed that it had disrupted five covert influence operations that attempted to misuse its models for deceptive activities across the internet.

Procreate Rejects Generative AI, Stands Firm in Support of Authentic Creativity

Procreate, an illustration and animation app, is resisting the trend of integrating generative AI features into its products. CEO James Cuda expressed his strong opposition to generative AI in digital art, emphasizing in a video titled "We’re never going there. Creativity is made, not generated" that such technology undermines true human creativity. In a post on X, he bluntly stated, "I really hate generative AI. I don’t like what’s happening in the industry, and I don’t like what it’s doing to artists." 

Cuda believes that by not adopting generative AI like its competitors, such as Adobe Photoshop, Procreate is staying true to its mission of supporting authentic creativity. While Procreate’s tools are compatible with Adobe Photoshop, the latter offers a broader range of AI-powered features, particularly for photo editing.

In contrast, Procreate’s rivals are fully embracing generative AI. Adobe has rolled out AI features through its Firefly models, while Canva has integrated AI tools via Magic Studio. 

Procreate’s stance isn't unique. In December 2022, Japanese illustration app Clip Studio Paint announced it would no longer introduce its AI image generator palette, citing concerns over exploiting artists’ intellectual property and a shift away from its core user base after receiving community backlash.

Elsewhere, the Hong Kong Monetary Authority has introduced new guidelines on using generative AI in consumer applications to protect end-users. These guidelines update the authority's four pillars of AI regulation in a circular directed mainly at banks.

Samsung to Begin Mass Production of Next-Gen HBM4 Memory by Late 2025 for AI GPUs

Samsung is reportedly set to finalize the design for its next-generation HBM4 memory in Q4 2024, with mass production anticipated by late 2025. The AI industry's growing demand for HBM memory has driven major producers like SK hynix, Samsung, and Micron to ramp up production, but it still hasn't met the surging needs. SK hynix is already fully booked for its HBM3 and HBM3E supplies through 2025, making the race to dominate HBM4 production crucial for AI GPU suppliers like NVIDIA and AMD.

NVIDIA’s upcoming Rubin R100 AI GPU is expected to utilize HBM4 memory, spurring both SK hynix and Samsung to invest heavily—spending billions of dollars and expanding semiconductor facilities in South Korea, Taiwan, and the U.S.—to meet the anticipated demand for HBM3, HBM3E, HBM4, and even HBM4E by 2025 and beyond.

According to The Elec, Samsung has begun preliminary work on HBM4 and is expected to complete its tape-out—when design and production methodologies are finalized—by Q4 2024. Sampling of this HBM4 memory will follow soon after, with companies like NVIDIA and AMD (especially NVIDIA) eager to benefit from the advancements.

Samsung plans to produce the logic dies for its HBM4 memory using its in-house 4nm process node. The logic die, a critical component in HBM chips for AI GPUs, sits at the base of the stacked memory dies.

While SK hynix, Samsung, and Micron currently lead in HBM production, the newer HBM4 memory demands a highly specialized foundry process with features tailored to the needs of AI chipmakers like NVIDIA, AMD, and Intel. Samsung's 4nm process, known for its 70%+ yield rate, is already used in its Exynos 2400 chipset, powering the latest Galaxy S24 smartphones.

Although Samsung has relied on 7nm technology since 2019, it now seems poised to shift to the more advanced 4nm node for its HBM4 logic dies. SK hynix, which has dominated the HBM market alongside NVIDIA in recent years, is also gearing up for HBM4 production in partnership with TSMC.

While NVIDIA’s Blackwell B100 and B200 AI GPUs will use the current HBM3E standard, the company’s future Rubin R100 AI GPU, expected in Q4 2025, will be equipped with ultra-fast HBM4 memory.

Survey Highlights Divided Opinions Among Game Developers on Generative AI

A recent survey conducted by the Game Developer Collective reveals that opinions among developers regarding generative AI remain divided. About 30 percent of respondents believe the technology will help cut development costs in the next year, while 8 percent think it will actually increase costs. Additionally, 35 percent remain skeptical, and 20 percent expect little to no impact.

Generative AI continues to spark debate both within and outside the gaming industry, and the survey indicates that this divide will persist. Critics argue against its environmental impact and claim that it produces lower-quality results compared to human artists. In terms of game quality, 17 percent of developers think genAI will enhance games in the coming year, 28 percent are indifferent, and 35 percent believe it will worsen game quality.

Concerns over generative AI have been raised particularly in relation to in-game art and voice acting. Since late July, SAG-AFTRA actors have been striking over the “existential threat” posed by genAI, including contracts that pressure them into giving up voice rights.

On the other hand, some argue that generative AI is essential for certain applications, like incorporating real-world athletes into EA Sports College Football 25. Studios like Nintendo haven’t entirely dismissed the technology but caution against using it as a quick fix during development.

Earlier this year, the Collective noted that 49 percent of surveyed developers reported their studios were already using generative AI, with another 31 percent utilizing it outside of work. Most of the in-office usage was related to financial tasks, community engagement, and production management.

TheOpensource.AI News

Linux Foundation Backs Open Model Initiative to Drive Open Source AI Innovation

The Linux Foundation, a prominent figure in the FOSS community known for its involvement in supporting around 900 open-source projects, providing developer training, and fostering a global developer network, is now turning its focus to the rapidly growing field of generative AI (GenAI) and the models that drive it.

Recently, they announced the launch of the Open Model Initiative (OMI), now officially hosted as an open-source foundation under The Linux Foundation. This new initiative aims to shape the future of open AI practices within GenAI by promoting open innovation in AI models.

The OMI was originally established by three AI-focused organizations—Invoke, Comfy Org, and Civitai—as a community-driven effort to encourage the development and use of openly licensed AI models for generating images, videos, and audio.

The initiative will be governed by a community-led Steering Committee. AI models released through this program will have open licenses that cannot be altered, will not include deletion clauses, and will avoid imposing recurring access fees on users.

Key goals of the initiative include:

- Establishing a governance framework and working groups to support collaborative community development.

- Conducting surveys to determine the open-source community's preferences for future AI model research and training.

- Supporting the creation of competitive, permissively licensed, and ethically developed AI models.

- Developing shared standards for model interoperability and consistent metadata practices.

- Creating transparent datasets for training and initiating captioning processes.

- Launching an alpha test model for focused Red Teaming efforts.

- Releasing an alpha version of the model with fine-tuning scripts by the end of 2024.

While OMI doesn't yet have an official website, information is available through its GitHub repository and Discord server.

During the announcement, Jim Zemlin, Executive Director of The Linux Foundation, emphasized the foundation's commitment to fostering open and collaborative AI development. He noted that the Open Model Initiative represents a major step toward making AI accessible and beneficial for all, enabling innovation without barriers.

Such an initiative is timely, given the increasing trend of organizations and individuals developing AI models primarily for financial gain, leading to situations where open access to models can be suddenly replaced by restrictive paid licensing arrangements.

How Defense Priorities Shape the Open-Source AI Debate?

Generative artificial intelligence (AI) has become a growing priority for the U.S. Department of Defense (DOD). AI-driven software using generative AI foundation models—versatile systems that mimic human reasoning—could be used to analyze vast amounts of intelligence data, automate Pentagon tasks, or enable autonomous navigation for aircraft, vehicles, and ships. Much of the progress in this area comes from commercial and academic research. If generative AI is widely adopted by the DOD, commercial foundation model developers will become an essential component of the defense industrial base. The Joint Force, therefore, has a vested interest in how the commercial foundation model ecosystem evolves.

The DOD’s AI strategies heavily rely on continued innovation from the commercial sector. To support this, the Pentagon has allocated new funding to acquire AI-powered technologies, such as for the Replicator drones and the Joint All-Domain Command and Control network, while establishing new entities like the Chief Digital and AI Office (CDAO) and Task Force Lima to manage these initiatives.

As these efforts progress, the Pentagon should consider the potential impacts of proposed regulations on the commercial foundation model market. Similar to how policies in spectrum auctions or shipbuilding affect defense options, civil sector regulations will influence future DOD decisions. Policies that promote a competitive ecosystem of players could strengthen the supply chain for future defense programs, while those that drive consolidation—like the Jones Act or the 1993 “Last Supper”—could hinder it. The 2023 Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence has already invoked the Defense Production Act to regulate the commercial AI market, raising questions about where market regulations could intersect with defense production.

Ongoing debates about AI safety, particularly regarding open foundation models, are of special interest. Unlike closed models like GPT-4, open foundation models such as Llama, Mistral, or Qwen openly share their underlying parameters (“weights”), allowing them to be inspected, modified, and used by others. With the performance of open models nearing that of their closed counterparts, some argue that their distribution could present significant risks. Others highlight the advantages of open models for research, security, and national competitiveness. While the outcome remains uncertain, proposals to restrict open model distribution—such as California Senate Bill (SB) 1047—are gaining legislative support.

The resolution of this debate on open foundation models will have direct consequences for the defense industrial base. As discussed later, there are early indications that a diverse open model ecosystem could be advantageous for the DOD by enhancing its ability to (1) competitively source and maintain AI systems, (2) securely deploy AI, and (3) address emerging use cases. The ongoing debate on open models serves as a case study for how civil society considerations might influence defense priorities in AI policy decisions.

Salesforce Unveils ‘xGen-MM’ Open-Source AI Models to Enhance Multimodal Understanding

Salesforce, a leader in enterprise software, has introduced a new collection of open-source large multimodal AI models aimed at speeding up research and development in more advanced AI systems. Known as xGen-MM (also referred to as BLIP-3), these models represent a major leap forward in AI’s capacity to process and generate content that integrates text, images, and other types of data.

In a research paper published on arXiv, Salesforce AI Research detailed the xGen-MM framework, which includes pre-trained models, datasets, and fine-tuning code. The largest model, with 4 billion parameters, performs competitively on various benchmarks when compared to other open-source models of similar size.

The paper’s authors emphasized that by making their models, curated large-scale datasets, and fine-tuning codebase openly available, they aim to advance large multimodal model (LMM) research. This open-source approach is a shift away from the trend of keeping sophisticated AI models proprietary, potentially making cutting-edge multimodal AI technology more accessible.

A standout feature of xGen-MM is its ability to handle “interleaved data,” which integrates multiple images and text—described by the researchers as “the most natural form of multimodal data.” This enables the models to carry out complex tasks, such as answering questions that involve analyzing multiple images simultaneously, which could be crucial in applications like medical diagnosis and autonomous vehicles.

The release includes different versions of the model, each optimized for specific purposes: a base pre-trained model, an instruction-tuned model for following directions, and a safety-tuned model designed to minimize harmful outputs. This range highlights the AI community’s growing recognition of the need to balance performance with safety and ethical considerations.

Salesforce’s decision to make these models open-source could greatly accelerate progress in the field. By providing high-quality models and datasets, they are allowing a broader group of researchers and developers to contribute to the development of multimodal AI. This approach contrasts with that of some tech companies that have kept their most advanced models private.

However, the release of such powerful models also raises important discussions about the risks and societal impacts of increasingly sophisticated AI systems. Despite the inclusion of safety measures, the broader consequences of making advanced AI models widely accessible remain a significant point of concern and debate within the tech community and beyond.

TheClosedsource.AI News

ChatGPT-4o Surpasses Google Gemini to Regain Top Spot in Chatbot Rankings

OpenAI's ChatGPT and Google Gemini have been competing fiercely for chatbot dominance, but the rivalry is intensifying. Although Claude briefly held the top spot on AI benchmarking platform LMSys Chatbot Arena earlier this year, Gemini had been leading the pack recently.

However, a new version of ChatGPT-4o (20240808) has now reclaimed first place with a score of 1314, surpassing Gemini-1.5-Pro-Exp by 17 points. This victory comes just a day after Google highlighted its lead on the arena board during its Made by Google event.

According to LMSys.org on X, the "New ChatGPT-4o shows significant advancements in technical fields, especially in Coding (with a 30+ point gain over GPT-4o-20240513), as well as in Instruction-following and Hard Prompts."

OpenAI had introduced the latest GPT-4o version in ChatGPT, with a similar model released for developers on the same day that the Chatbot Arena results were published. 

In our own testing, we noticed this version was noticeably faster than previous ones, even managing to build an entire iOS app within an hour. Combined with recent enhancements to the Mac app, it's been an especially significant week for ChatGPT users and OpenAI.

With new models and upgrades launching regularly, there’s a good chance we’ll see further changes in the leaderboard in the coming weeks or months. Meanwhile, Google Ultra 1.5 and Claude Opus 1.5 are yet to debut, and xAI's Grok 2 has already made its way into the top ten.

Eric Schmidt Bets on NVIDIA, Calls Microsoft’s OpenAI Deal a “Stupid” Move

Every major tech company is rushing to join the AI trend. While Microsoft, OpenAI, and Google have found some success, former Google CEO Eric Schmidt recently suggested that NVIDIA could emerge as the biggest winner in the AI race (via CNBC).

Schmidt noted that Google may have lost its AI lead due to its work-from-home policy, which allowed startups like OpenAI and Anthropic to gain an advantage. "The startups succeed because their people work incredibly hard," Schmidt added.

Paul Buchheit, the developer behind Gmail, echoed this view but pointed to Alphabet’s reorganization, which placed Sundar Pichai as CEO, as a turning point. According to Buchheit, this shift caused the company’s AI ambitions to stall as its focus turned to maintaining dominance in search.

Despite Microsoft’s significant investment in OpenAI, Schmidt believes NVIDIA holds the advantage in AI. He pointed out that most tech companies jumping into AI depend heavily on NVIDIA’s chips.

“I’m talking to the big companies, and they’re telling me they need $20 billion, $50 billion, $100 billion — it’s really tough,” Schmidt explained. This suggests that much of the funding for AI projects is being funneled toward NVIDIA for its AI chips.

While the stock market remains volatile, NVIDIA’s market valuation has surged, making it the world’s most valuable company, surpassing Apple and Microsoft with a market cap of $3.335 trillion. Last year, NVIDIA was ranked as the most profitable chip brand, with clients like Microsoft and Meta.

"If $300 billion is going to Nvidia, you know what to do in the stock market. That’s not financial advice."

Schmidt also mentioned that NVIDIA won’t be the only company benefiting from AI investments. Other major players capable of investing in NVIDIA and its data centers will also be well-positioned in the AI race.

Regarding Microsoft’s partnership with OpenAI, Schmidt admitted it’s a complex relationship. He initially thought the deal was a poor decision, saying, "When Microsoft partnered with OpenAI, I thought it was the dumbest idea ever, effectively outsourcing their AI leadership to OpenAI and Sam Altman’s team. But now they’re on track to become the most valuable company." This comes as investors expressed concerns over Microsoft’s heavy investment in AI with limited immediate profit returns.

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!