• Posts
  • OpenAI Releases ‘Deepfake’ Detector to Disinformation Researchers

OpenAI Releases ‘Deepfake’ Detector to Disinformation Researchers

Also: Singapore and US to explore AI and other dual-use tech in new defence innovation agreement

Good morning,

In today’s newsletter, we delve into OpenAI's launch of a high-accuracy deepfake detector to aid disinformation researchers, Singapore's Project Moonshot for generative AI safety, and McKinsey's report on the spike in generative AI adoption generating significant business value. Additionally, we cover the new US-Singapore MOU on defense innovation focusing on AI, Google's temporary rollback of its AI Search feature due to inaccuracies, and Foxconn's strategic pivot to electric vehicles and AI servers for growth. We also highlight Microsoft's AI Copilot integration into Telegram and the trend of VCs selling shares of leading AI companies to smaller investors via SPVs.


  • ⚠️ OpenAI Releases ‘Deepfake’ Detector to Disinformation Researchers

  • 🇸🇬 Singapore rolls out new toolkit to test gen AI safety

  • 🧐 McKinsey The state of AI in early 2024: Gen AI adoption spikes and starts to generate value

  • 🪖 Singapore and US to explore AI and other dual-use tech in new defence innovation agreement

  • 😐 Google Rolls Back A.I. Search Feature After Flubs and Flaws

  • 📈 Apple assembler Foxconn turns to EVs and AI servers for next growth

OpenAI has introduced a deepfake detector to assist disinformation researchers in identifying AI-generated content, particularly from its image generator DALL-E 3. This detector, which boasts a 98.8% accuracy rate for images created by DALL-E 3, aims to support efforts in mitigating the influence of deepfakes, especially with upcoming major elections worldwide. However, OpenAI acknowledges the tool's limitations, as it is not effective against content from other generators like Midjourney. The initiative is part of a broader industry effort, including collaborations with Google and Meta under the Coalition for Content Provenance and Authenticity (C2PA), which seeks to develop digital content credentials to trace the origins of AI-manipulated media. OpenAI is also exploring watermarking techniques for AI-generated audio to enhance traceability and combat misuse.

Singapore has launched Project Moonshot, an ambitious initiative to test the safety of generative AI models, addressing security challenges linked to large language models (LLMs). Developed by the AI Verify Foundation and the Infocomm Media Development Authority (IMDA) in collaboration with partners like DataRobot and Temasek, this open-source toolkit helps AI app developers assess their applications against specific benchmarks and includes red-teaming modules to test resilience against manipulation. Project Moonshot aims to balance AI innovation with robust governance, supporting businesses while safeguarding citizens. Additionally, Singapore is redesignating its Digital Trust Centre as the AI Safety Institute, part of a global network focused on AI governance. The country is also working on establishing a common testing benchmark for LLMs in collaboration with MLCommons and the AI Verify Foundation to enhance safety and trustworthiness in AI applications.

Generative AI adoption has surged significantly in 2024, with 65% of organizations now regularly using it, nearly double the adoption rate from ten months ago. This rapid increase reflects growing confidence in gen AI's ability to drive business value, evidenced by cost reductions and revenue increases reported by users. The most common applications are in marketing, sales, product development, and IT. Despite the benefits, gen AI also presents risks such as inaccuracy and data privacy issues. Leading organizations, which are already attributing significant portions of their EBIT to gen AI, are implementing best practices like risk mitigation and customization of AI models to maximize value. Investment in gen AI continues to grow, with many companies dedicating substantial portions of their digital budgets to both generative and analytical AI.

Singapore and the United States have signed a memorandum of understanding (MOU) to enhance cooperation on defence innovation, focusing on AI and autonomous systems. This agreement, signed at the Shangri-La Dialogue 2024, aims to address operational challenges faced by both militaries and develop technologies for counter-drone capabilities and maritime security. The MOU, signed by Singapore’s chief defence scientist Tan Peng Yam and Douglas Beck from the US Defence Innovation Unit, reflects a commitment to integrating emerging dual-use technologies into defence strategies. US Secretary of Defence Lloyd Austin and Singapore Defence Minister Ng Eng Hen, who witnessed the signing, emphasized the importance of leveraging commercial technologies for defence purposes. Both nations also reaffirmed their longstanding defence relationship and discussed regional security issues and geopolitical developments, underscoring the importance of US engagement in the Asia-Pacific region.

Google has temporarily reduced the presence of its AI Overviews feature in search results due to a series of errors and misinformation. Introduced by CEO Sundar Pichai, the feature aimed to generate concise, informative summaries for search queries. However, it faced criticism after producing inaccurate and sometimes dangerous advice, such as suggesting glue for a pizza recipe and recommending the ingestion of rocks for nutrients. In response, Google implemented refinements to minimize such mistakes, particularly in health-related queries and content from forums. Despite these setbacks, Google continues to develop AI Overviews and remains committed to improving the feature. This move highlights the delicate balance Google must strike between integrating advanced AI technologies and maintaining the reliability of its search engine amidst competitive pressures from Microsoft and OpenAI.

Foxconn, the Taiwanese company known for assembling Apple's iPhones, is diversifying its operations to address slowing smartphone market growth and low profit margins. The company is focusing on electric vehicles (EVs) and AI servers to drive future growth. It plans to partner with Japanese automakers for the mass production of EVs, including the Foxtron Model B, and expand its presence in the Southeast Asian and North American markets. Foxconn is also developing next-generation silicon carbide semiconductors and EV batteries. Concurrently, it is enhancing its AI server business, capitalizing on orders from Nvidia and major U.S. cloud companies. This pivot aims to replicate its successful contract production model, expecting AI server-related sales to potentially rival its iPhone-related sales by 2027-2028. Additionally, Foxconn's acquisition of Sharp involves restructuring efforts to streamline operations.

🛠️ AI tools updates

Microsoft's AI, Copilot, is now available on Telegram in its beta phase, allowing users to interact with the AI directly within the messaging app. This integration provides Telegram users with easy access to Copilot's chat capabilities without needing additional applications. While this beta version may have some limitations, such as potential hallucinations and text-only responses, it signifies a strategic collaboration that benefits both Microsoft and Telegram. Microsoft's initiative reflects its commitment to AI development, enhancing user experience and expanding the reach of its AI services. This move also aligns with the growing trend of integrating AI into widely used platforms, making advanced AI tools more accessible to a broader audience.

💵 Venture Capital updates

VCs are increasingly utilizing special purpose vehicles (SPVs) to sell shares of leading AI companies like Anthropic, Groq, OpenAI, Perplexity, and xAI to smaller investors, including family offices and high-net-worth individuals. This practice allows investors who might not otherwise have access to these high-demand stocks to participate by pooling resources. SPVs, formed by those with direct access to the shares, often charge significant fees and retain some profit share. This trend highlights the growing interest and competitive nature of investing in AI startups, as well as the complex and sometimes opaque nature of SPV arrangements, which can vary significantly in terms and conditions for investors.

🫡 Meme of the day

⭐️ Generative AI image of the day