AI in Marketing

Digital billboards are reshaping the out-of-home advertising landscape, offering dynamic creative opportunities previously unseen with static displays. From 3D images to full-motion graphics, digital billboards allow advertisers to engage audiences in new ways. The Out of Home Advertising Association of America reports that the number of digital billboards in the U.S. has nearly doubled from 16,000 in 2023 to about 30,000 today, reflecting a growing shift towards digital in the industry. This shift is influencing media buying and planning strategies as well as inspiring innovative creative approaches. In 2017, Coca-Cola made headlines with a 3D billboard in Times Square, illustrating the creative potential of this medium. Digital billboards can be more resource-intensive than static ones, but their location in high foot-traffic areas like Times Square and the Las Vegas Strip makes them attractive to brands aiming for high visibility. John Mergen, chief media officer at The Shipyard, highlights the captivating nature of digital billboards, noting their ability to draw attention and engage passersby. While digital billboards still represent a smaller fraction of the 350,000 static billboards in the U.S., their increasing presence signifies a gradual but certain transformation of the advertising industry.

From digital billboards to AI-driven brand strategies, the advertising world evolves.

WPP and Google have announced a five-year partnership expansion, with WPP committing $400 million towards Google technologies. This deal, finalized in Mountain View, California, involves WPP’s CEO Cindy Rose and Google Cloud CEO Thomas Kurian. The collaboration aims to revolutionize brand strategies in creative, media, and commerce through AI-driven solutions. WPP will integrate Google’s AI into its services, including bespoke AI models and secure data collaboration. The partnership also focuses on enhancing video and image production and developing new AI talent through WPP's expanded Creative Technology Apprenticeship program, with Google as the main curriculum partner. This strategic move is expected to accelerate marketing innovation and drive significant growth for both companies.

Turning to Andrej Karpathy's nanochat, we see AI's coding challenges.

OpenAI co-founder Andrej Karpathy, who famously coined "vibe coding," recently acknowledged its limitations. After leaving OpenAI, he introduced his open-source model, nanochat, which he entirely hand-coded due to AI tools' inadequacies. Nanochat aims to enable anyone to build a ChatGPT-style chatbot swiftly and affordably. Karpathy's admission contrasts with his previous stance that vibe coding—using AI for coding without scrutinizing the output—suffices for simple projects. However, nanochat’s complexity required precise manual coding. A survey by Fastly supports this, revealing that 95% of developers spend extra time correcting AI-generated code. Research by METR indicates that AI tools can slow down task completion, prompting some companies to hire specialists for error correction.

On the development front, Google's Veo 3.1 enhances AI video innovations.

Google has announced the release of Veo 3.1, an upgrade to its AI video model, enhancing realism and prompt accuracy. Available through the Gemini app and Flow filmmaking tool, Veo 3.1 introduces a "Fast" variant for cost-effective video generation. The model offers improved audio, supports both landscape and portrait 16:9 formats, and aligns with the growing trend of vertical videos, potentially impacting platforms like YouTube Shorts and TikTok. Developers can access Veo 3.1 through Vertex AI and the Gemini API, benefiting from new capabilities in video creation. These include integrating custom audio with visual elements. Veo 3.1 positions Google competitively against rivals like OpenAI in the AI video domain.

In research terms, MIT's study challenges the AI model scaling trend.

A recent study from MIT suggests that the largest and most computation-heavy AI models may soon hit a point of diminishing returns, where further scaling brings fewer benefits compared to smaller, more efficient models. Researchers have found that while giant models face challenges in achieving significant performance gains, smaller models running on less powerful hardware could become increasingly effective in the next decade. Neil Thompson, a computer scientist at MIT, predicts a narrowing of capabilities within five to ten years. This finding presents a reality check for the AI industry, which has traditionally relied on vast computational resources. Notably, efficiency improvements, like those seen with DeepSeek’s cost-effective model, highlight the potential for smaller models to compete strongly. Hans Gundlach, an MIT research scientist, points out that this trend is particularly evident in reasoning models, which require more computation during inference. Thompson emphasizes the importance of developing more efficient algorithms alongside scaling up compute power. The study emerges amid a boom in AI infrastructure investment, with companies like OpenAI committing billions to expand in the U.S. Despite the industry's push for more computing power, this research suggests that focusing on algorithm efficiency might be a more sustainable and impactful path forward.

Veo 3.1's advancements highlight creative control, aligning with user feedback.

Google has rolled out Veo 3.1, its newest AI video generation model, enhancing creative and technical features like narrative control and audio integration. This release opens up new possibilities for enterprises and developers seeking scalable video tools. The model offers improved physics and quality without price changes and includes robust editing features. Veo 3.1 supports dialogue and ambient sound, integrating native audio generation in key features of Google’s Flow. This development eliminates the need for post-production audio work, streamlining content creation for businesses. Users can now employ various input types, including text prompts and images, for more control over outputs, allowing seamless scene transitions. Feedback from users drove these upgrades, emphasizing the importance of artistic control.

From Google's Flow to Anthropic's Claude Haiku 4.5, AI tools diversify.

Anthropic is offering its Claude Haiku 4.5 AI model for free, aiming to challenge OpenAI's dominance in the AI market. This model, priced at $1 per million input tokens and $5 per million output tokens, is cheaper and faster than its predecessor, Claude Sonnet 4. It democratizes access to advanced AI capabilities, previously limited to premium models, allowing free users of Claude.ai to benefit. Haiku 4.5 is designed to operate autonomously on computers and works in multi-agent systems, enabling complex tasks like software refactoring. This approach lets enterprises use Sonnet 4.5 for planning and Haiku 4.5 for execution, mimicking human organizational structures and enhancing efficiency in complex projects.

Amid Anthropic's innovations, concerns over AI risks are increasingly discussed.

The New York Times reports that concerns over the existential risks posed by artificial intelligence are gaining traction, with AI researcher Eliezer Yudkowsky expressing grave warnings about the potential for catastrophic outcomes. The discourse on AI's potential to displace or even destroy humanity intensified following the release of ChatGPT, as experts began contemplating the probability of a rogue AI scenario. In 2023, prominent figures like Sam Altman, Bill Gates, and Geoffrey Hinton endorsed a statement urging that mitigating AI extinction risks be prioritized alongside threats like pandemics and nuclear warfare. Despite these warnings, advancements in AI technology have continued, with companies prioritizing financial growth over existential concerns. Yudkowsky, an early proponent of AI caution, recently co-authored a book titled “If Anyone Builds It, Everyone Dies,” aiming to alert the broader public about the dangers he foresees. His longstanding involvement in the AI community has influenced numerous developers, though he struggles to deter them from continuing their work on potentially perilous technology. The conversation about AI's potential threats remains contentious and complex. While some experts take the risks seriously, others, including host Ezra Klein, maintain skepticism about the likelihood of AI takeovers, despite acknowledging the profound implications of developing superintelligent systems. During a recent episode of “The Ezra Klein Show,” Yudkowsky elaborated on his perspective that AI development is akin to nurturing a growing entity, rather than merely crafting a controllable tool. This analogy underscores his belief in the inherent unpredictability and danger of AI systems as they evolve. As the debate continues, Yudkowsky's efforts to raise awareness highlight the urgent need for reflection on the trajectory of AI technology and the societal priorities that shape its future.