Good morning, it’s Thursday! What do Honolulu’s permits, San Francisco’s traffic lights, and wildfire detection in California have in common? AI is quietly revolutionizing cities, streamlining red tape, optimizing urban systems, and even aiding in disaster prevention. Yet, despite these breakthroughs, U.S. cities lag in global "smart city" rankings. So, what’s holding us back?
🗞️ Top Stories: AI Revolutionizes Weather Forecasting, Threatens Music Industry, Turns Images Into 3D Worlds, and More
🌆 AI Cities: How U.S. Cities Are Using AI to Transform Governance
👾 FF Original: Econ 06 | How Software Ate the World
🤖 Machine Learning: Unlocking Insights from Sparse Data
📗 Research Papers: Swift Solutions for AI Jailbreaks Show Promise
📂 Open-Source: Video Generation Model Sets New Benchmark
📽️ Video: Text-to-Video Model Locally Tutorial (Mochi-1)
🧰 Tools: AI Tools to Summarize, Recolor, and Plan
🗞️ YOUR DAILY ROLLUP
Top Stories of the Day
🌦️ AI Revolutionizes Weather Forecasting with GenCast
Google DeepMind's GenCast outperforms the ECMWF's leading ensemble system on 97.2% of key metrics, particularly excelling in extreme weather predictions and cyclone tracking. Powered by advanced ML, it generates forecasts in just 8 minutes—dramatically faster than traditional methods. While not a complete replacement for physics-based models, GenCast demonstrates significant potential in tackling the challenges of climate change.
🕹️ AI Transforms Images into Playable 3D Worlds
Google DeepMind is on a roll today… This time, it’s Genie 2, a tool that conjures up interactive 3D environments from single images—complete with consistent scenes, realistic physics, and NPC interactions. Playable for up to a minute, it pairs perfectly with DeepMind’s SIMA agent for cutting-edge AI training. While still in the research phase, Genie 2 teases a future where AI reshapes game design and blurs the lines between virtual and real worlds.
🎵 AI Threatens Music Industry Incomes by 25%
A CISAC study forecasts a 25% income drop for music industry workers by 2028 as generative AI reshapes the market. AI-generated music may claim 20% of streaming revenue, reducing creator opportunities. CISAC urges global action to regulate AI, emphasizing Australia and New Zealand’s proactive policies to safeguard creativity and mitigate cultural and economic impacts.
📱 Apple Faces Challenges Adapting Baidu AI for China
Apple's efforts to integrate Baidu's Ernie 4.0 AI into Chinese iPhones face hurdles in prompt accuracy and privacy policy conflicts. Apple's data restrictions clash with Baidu's need for interaction analysis, delaying progress. With Huawei's AI-driven phones gaining market share, the absence of advanced AI in iPhone 16 risks further sales declines in China, Apple's largest smartphone market.
🌆 AI CITIES
From Red Tape to Smart Solutions: How U.S. Cities Are Leveraging AI to Govern
The Recap: U.S. cities are leveraging AI to streamline bureaucracies, make data-driven decisions, and engage communities, but entrenched barriers like rigid regulations, siloed operations, and risk-averse leadership slow adoption. The solution lies in strategic vision, fostering innovation, and ethical AI governance to maximize AI's potential for civic transformation.
Rigid bureaucracies, restrictive regulations, and leadership risk aversion are the primary obstacles preventing AI adoption in local governments.
Cities like Honolulu have cut bureaucratic inefficiencies, with AI expediting construction permitting by 70%, saving time and money without replacing human workers.
AI enables cities to optimize traffic, forecast crises, and allocate resources dynamically, as seen with wildfire detection in California and traffic improvements in Seattle.
AI tools like chatbots simplify administrative interactions, while platforms like New Rochelle’s urban modeling empower citizens to shape local policies.
Cities achieving AI success align projects with specific goals, de-bottleneck bureaucratic hurdles, foster public-private partnerships, and adhere to strict ethical AI principles.
Partnerships, like Columbus' “Smart Columbus Initiative,” enable cities to leverage expertise and resources from private companies and academic institutions.
Ethical oversight, inclusivity, and privacy protections are vital to ensure AI systems are fair, transparent, and aligned with community needs.
Forward Future Takeaways:
AI offers a transformative opportunity for American cities to enhance efficiency, transparency, and citizen engagement, potentially reshaping urban governance. However, success hinges on visionary leadership willing to embrace innovation and break bureaucratic silos while adhering to ethical governance principles. With thoughtful implementation, cities can not only improve local services but also set a global example for sustainable urban AI deployment. → Read the full article here.
👾 FORWARD FUTURE ORIGINAL
How Software Changed Everything - The Economics Behind Digital Dominance
In the previous article, we saw how progress in services automation got ahead of goods thanks to the rapid development of software engineering and the special characteristics and advantages of the digital realm compared to the physical. To get a fuller picture of why this matters in the current AI and Economics context, let’s dig deeper into, well, how software ate the world.
We touched upon these key differences and advantages: software, both data and code, is infinitely replicable (subject to underlying hardware constraints, but that’s usually not a problem these days for most cases). It is easy to experiment with software without the cost and risks to analogous experimentation in the physical world.
In the digital world, one can work remotely and collaborate from anywhere. And finally, the creative-commons free market of open-source development is the cherry on top! These and related unique characteristics and factors have enabled unprecedented economic transformation. Let’s see how:
Software has zero marginal cost. There are high fixed costs - the equipment, application development, etc. But then there’s near-zero distribution costs. It’s true that software development is typically an incremental process, but this is by and large for new features, and for fixing any issues. But it costs practically nothing to distribute another copy of the same version of software. By contrast, there is a real cost of manufacturing another physical widget of the exact same specification. → Continue reading here.
🤖 MACHINE LEARNING
AGCL: A Privacy-First Machine Learning Model Revolutionizing Sparse Data Insights
The Recap: A new machine learning framework called Attributed Graph Contrastive Learning (AGCL), developed by Jiyeon Hong and collaborators, enables organizations to make better use of sparse consumer data. Tested on donors with minimal interaction histories, AGCL offers privacy-conscious, actionable insights that improve recommendations and engagement strategies.
AGCL synthesizes minimal data from donors with limited demographic or interaction histories, enabling actionable insights.
The model increased success in project recommendations from 24% (traditional methods) to 34%.
AGCL combines donor-project interactions, donor connections, and project similarities to create a "neighborhood" of attributes, uncovering latent interests.
The model works without extensive data collection, aligning with growing concerns about data privacy.
AGCL maps donor preferences into "interest clusters," aiding marketers in crafting targeted strategies.
Beyond donor insights, AGCL could revolutionize customer segmentation by uncovering deep connections from sparse data.
AGCL empowers organizations with limited data resources to achieve personalized, impactful engagement.
Forward Future Takeaways:
AGCL represents a significant leap in leveraging minimal data for meaningful consumer insights, especially for non-profits and privacy-conscious organizations. Its ability to enhance personalization without compromising data privacy could establish it as a standard in donor and customer segmentation. As this technology scales, it may reshape marketing by enabling even small entities to deliver tailored experiences that build enduring connections. → Read the full article here.
🛰️ NEWS
Looking Forward
☁️ AWS Enhances Bedrock: With new caching and prompt routing features that cut costs up to 90%, and speed responses by routing simpler queries to smaller models. A marketplace for niche LLMs debuts too!
🎧 Spotify Wrapped 2024 Adds AI Podcast: This year’s recap includes an AI-powered podcast using Google’s NotebookLM dissecting your music habits alongside enhanced AI Playlists and DJ.
📝 Krea AI Unveils Krea Editor: The new tool promises a magical editing experience. Beta access is open for creators eager to explore its transformative features!
🔬 RESEARCH PAPERS
Rapid Defense Against AI Jailbreaks: New Techniques Show Promise
Researchers have introduced a novel approach to combat jailbreak attacks on large language models (LLMs) by focusing on rapid response techniques instead of striving for invulnerability. Using RapidResponseBench, a benchmark for adapting defenses to observed attacks, they developed a method that generates similar jailbreaks to train models for better resistance.
Their best-performing strategy reduced attack success rates by over 240x on known jailbreak types and 15x on new ones, even after observing just one example of each strategy. These findings underscore the potential of rapid adaptation to mitigate misuse of LLMs effectively. → Read the full paper here.
📂 OPEN SOURCE
HunyuanVideo: Open-Source Video Generation Model Sets New Benchmark
HunyuanVideo, a new open-source video generation framework from the team behind Hunyuan-Large, leverages advanced AI techniques to outperform leading closed-source models like Runway Gen-3 and Luma 1.6. Combining a unified image-video generative architecture, 3D VAE compression, and an MLLM text encoder, it offers high-quality video generation with superior motion, visual fidelity, and text alignment.
With over 13 billion parameters, HunyuanVideo supports both image-to-video and text-to-video capabilities, enhanced by its novel "prompt rewrite" feature for more accurate and visually appealing outputs. By open-sourcing the framework, the creators aim to democratize access to cutting-edge video generation tools. → Check it out on GitHub.
📽️ VIDEO
Text-to-Video Model LOCALLY Tutorial (Mochi-1)
Genmo AI’s Mochi 1 enables text-to-video generation on local machines using open-source tools like ComfyUI. This tutorial demonstrates how to set up the model, create videos, and customize settings on high-end hardware. Despite some limitations, Mochi 1 showcases impressive results, offering a glimpse into the future of accessible AI-powered video creation. Get the full scoop in our latest video! 👇
Reply