Dark Forecast For Media in AI Era, 2024 Overall Progress Remarkable

⚡️ Headlines

🤖 AI

Microsoft works to add non-OpenAI models into 365 Copilot products – Microsoft is integrating its own and third-party AI models into its 365 Copilot suite to reduce reliance on OpenAI's technology and enhance performance for enterprise users. [Reuters]

60 of our biggest AI announcements in 2024 – Google recaps its major AI developments in 2024, including the launch of Gemini, advancements in NotebookLM, and other significant AI milestones. [Google Blog]

Don't Look Now, but China's AI Is Catching Up Fast – Chinese startups are rapidly advancing in AI technology, closing the gap with American leaders despite facing limitations on accessing advanced AI chips. [The Wall Street Journal]

🦾 Emerging Tech

A Looming Threat to Bitcoin: The Risk of a Quantum Hack – The rise of quantum computing poses a potential threat to Bitcoin's security, as future quantum hacks could undermine the cryptocurrency's foundational cryptographic protections. [The Wall Street Journal]

🤳 Social Media

Apple vs Android: Does Uber charge more for iPhone users? Social media post starts discussion – A social media post has sparked a debate over whether Uber charges higher fares for iPhone users compared to Android users, raising questions about pricing algorithms. [Mint]

Blogs to Bluesky: social media shifts responses after 2004 tsunami – An analysis of how social media platforms have transformed disaster response and information dissemination since the 2004 tsunami, highlighting the evolution from blogs to platforms like Bluesky. [Tech Xplore]

🔬 Research

An AI system has reached human level on a test for 'general intelligence'. Here's what that means – OpenAI's latest model, o3, has achieved human-level performance on the ARC-AGI benchmark, indicating significant progress toward artificial general intelligence. [The Conversation]

Why Do Task Vectors Exist in Pretrained LLMs? This AI Research from MIT and Improbable AI Uncovers How Transformers Form Internal Abstractions and the Mechanisms Behind In-Context Learning (ICL) – Researchers from MIT and Improbable AI explore the existence of task vectors in pretrained large language models, shedding light on how transformers develop internal abstractions and mechanisms that facilitate in-context learning. [MarkTechPost]

⚖ Legal

Animation Guild Ratification Vote on Artificial Intelligence – The Animation Guild is set to hold a ratification vote concerning the use of artificial intelligence in animation, addressing industry concerns over job security and creative control. [Variety]

🎱 Random

Telegram Closes 2024 With $1B in Revenue and Its First Profit, Says CEO Pavel Durov – Telegram reports surpassing $1 billion in revenue and achieving profitability for the first time in 2024, marking a significant milestone for the messaging platform. [Observer]

🔌 Plug-Into-This

Dark Musings on Media & AI

AI platforms are increasingly seen as potential revenue sources for publishers through licensing deals, but these arrangements often provide only short-term financial relief while failing to address deeper structural challenges in the media industry.

• Short-Term Nature of Licensing Deals: AI companies are entering agreements with publishers, offering immediate cash infusions that fail to deliver sustainable revenue or address declining ad revenues.

• Limited Value for Publishers: Licensing deals often benefit AI platforms more than publishers, as platforms gain access to vast amounts of content while publishers struggle to monetize effectively.

• Underlying Industry Challenges Persist: Issues like diminishing advertising revenue, declining subscriptions, and competition with free online content remain unresolved despite AI-driven partnerships.

• Risk of Dependency on AI Platforms: Relying on AI partnerships could further weaken publishers' independence and bargaining power, making them reliant on tech companies for survival.

• Need for Long-Term Solutions: Sustainable revenue models and stronger protections for original content are essential to prevent further erosion of the media ecosystem.

📉 AI licensing deals may provide temporary financial relief but risk exacerbating the imbalance between tech platforms and publishers in the fight for digital relevance and sustainability.

2024 Brought Some Insane Progress in AI...

The rapid advancement of artificial intelligence in recent months has seen breakthroughs in model performance, accessibility, and applications, ushering in a new era of AI capabilities. From near-universal availability of GPT-4-class models to the emergence of reasoning-focused models like o1, these innovations are reshaping fields from research to real-time interactions.

• Explosion of Advanced Models: GPT-4-class AIs are now widespread, with open-source and localized models enabling high-level AI functions on personal devices, democratizing access.

• Introduction of Gen3 AI: The next generation of models, including o1, exhibits advanced reasoning abilities, capable of complex problem-solving and even generating new research ideas.

• AI in Research Applications: Tools like o1-pro are collaborating with researchers, offering novel approaches to problems, though expert oversight remains critical for evaluating outputs.

• Real-Time Multisensory Interaction: Innovations like Gemini 2.0 allow AIs to process visual and auditory input, creating opportunities for immersive and interactive applications.

🤝 The hype was real I guess! Back in January 2024, you could have been forgiven to assume that progress wouldn’t match the lofty claims being thrown around in the AI space, but we certainly have seen widespread availability of truly advanced AI hitting the mainstream.

OpenAI Trained o1 and o3 to Think About Its Safety Policy

OpenAI has enhanced o1 and o3 with advanced deliberative alignment techniques, enabling them to reason about safety policies as part of their decision-making processes. This marks a significant step in integrating ethical considerations directly into AI reasoning.

• Deliberative Safety Integration: o1 and o3 are trained to assess safety rules contextually, improving adherence to ethical guidelines while performing complex tasks.

• Advances in Reasoning Performance: The models demonstrate exceptional capabilities in coding, mathematics, and other intricate domains, surpassing prior iterations in both accuracy and problem-solving depth.

• Controlled Release for Safety Testing: OpenAI is limiting access to these models for now, conducting rigorous safety evaluations and inviting researchers to participate in early testing.

• Focus on Mitigating Risks: These developments reflect heightened attention to AI safety, ensuring that the models' advanced functions do not lead to unintended or harmful outcomes.

• Balancing Power and Responsibility: OpenAI continues refining o1 and o3 to align cutting-edge AI performance with strong safeguards, ensuring responsible use in real-world applications.

😇 If this methodology works well it could serve as a blueprint for the rest of the industry in developing “ethical AI” — whatever that ends up meaning.

 🆕 Updates

Meta is reportedly planning to release an updated version of its Ray-Ban smart glasses in 2025, featuring an integrated display capable of showing notifications and responses from Meta's AI assistant.

Why Care?

• Enhanced Functionality: The addition of an in-lens display will allow users to view notifications and AI responses directly through the glasses, reducing reliance on smartphones.

• Seamless Integration: Combining eyewear with digital information aims to provide a more intuitive and hands-free user experience.

• Competitive Edge: This development positions Meta as a strong contender in the wearable tech market, competing with other tech giants investing in smart glasses.

• Future of Computing: Aligns with Meta CEO Mark Zuckerberg's vision of smart glasses becoming the next major computing platform.

• User Convenience: Potential to streamline daily tasks by providing immediate access to information without the need to check other devices.

Kling AI, developed by Kuaishou Technology, has recently released version 1.6

This update introduces significant enhancements in video generation capabilities, including improved prompt responsiveness, visual fidelity, and realism in physical interactions. Users can now create more lifelike animations with subtle emotional expressions and complex movements, resulting in high-quality, cinema-grade videos up to two minutes long. Additionally, Kling AI 1.6 offers better semantic understanding, enabling the generation of coherent narratives and dynamic scenes. The platform remains accessible with an affordable pricing model, charging approximately 35 credits for every five seconds of generated video, with free credits available upon signup.

Why Care?

• Enhanced Realism: Experience more lifelike animations with improved physical interactions and emotional expressions.

• Extended Video Length: Create longer videos, up to two minutes, maintaining high-quality output at 30 frames per second.

• Improved Prompt Responsiveness: Generate coherent narratives and dynamic scenes with better semantic understanding.

• Affordable Access: Benefit from a cost-effective pricing model with free credits upon signup.

• User-Friendly Interface: Enjoy an intuitive platform designed for both novice and experienced content creators.

📽️ Daily Demo

🗣️ Discourse