Welcome to our second AI Trailblazers Newsletter. Our approach diverges from standard newsletters. Instead of just summarizing AI news, we delve into its takeaways for your business, career, and society, while also sharing specific AI Trailblazer highlights.

Thank you for your early feedback. We continue to enhance the newsletter based on community input. If you find this valuable, please share it with your peers. We welcome your feedback, too. Please reply or reach out directly.

Mysterious LLMs: How Today's Models Defy Human Understanding

The MIT Technology Review explored the astounding capabilities of large language models, which accomplish feats that are not fully understood by humans. The article considers why this lack of understanding could pose a problem. Highlights include:

• Unpredictable Learning Phenomena: Large language models (LLMs) like OpenAI's GPT-4 exhibit unexpected behaviors, such as "grokking," where models suddenly understand a task after prolonged training, challenging our understanding of how deep learning works. Researchers are exploring these phenomena without a clear consensus on their causes, indicating a gap in our understanding of AI's learning processes.

• Beyond Classical Statistics: LLMs defy expectations set by classical statistics, showcasing abilities to generalize and perform tasks in ways not predicted by existing mathematical frameworks. This includes solving problems in languages not directly trained on and displaying capabilities that emerge unexpectedly, suggesting LLMs operate on principles beyond our current theoretical grasp.

• The Double-Descent Phenomenon: Contrary to statistical predictions that larger models should eventually perform worse due to overfitting, research shows a "double descent" behavior where error rates decrease again as models grow, suggesting that LLMs can overcome overfitting and improve with size, challenging existing theories of model training and complexity.

• Implications for Future AI Development and Safety: Understanding the underlying mechanisms of LLMs is not just academically interesting but crucial for developing more efficient and capable AI systems and managing potential risks. As AI models become more powerful, the need for a solid theoretical foundation grows, not only to harness their potential but also to ensure they can be controlled and aligned with human values.

AI Trailblazer Takeaways: For business leaders, this means aggressively leveraging AI's potential to drive efficiency and creativity must be balanced with a cautiousness by instituting appropriate checks as knowing how LLMs may respond to specific business situations can be difficult to predict at times. Human oversight is essential.

AI in the Workplace: Unseen Impacts

Two professors from Northeastern University presented new research in the Harvard Business Review, showing the unexpected and unintended consequences of integrating AI into the workplace. Highlights below.

• Impact on Creative Perception: Research indicates that employees managed by AI algorithms are perceived as less creative by leaders, challenging the expectation that AI will enhance human creativity and innovation in the workplace.

• Reduced Investment in Innovation: Teams where their performance is overseen by algorithms receive less funding for innovation projects, suggesting that algorithmic management may inadvertently hinder a company's innovative capacities.

• Experiment Findings: Experiments show that regardless of the prevalence of algorithmic management within a company, employees managed by AI are viewed as less creative and are allocated fewer innovation resources.

• Recommendations for Leaders: To prevent AI from diminishing employees' creative potential, leaders should ensure algorithmic management focuses on routine tasks and not on areas requiring human creativity and emotional intelligence.

• Broad Implications for AI Deployment: The findings emphasize the importance of careful and deliberate AI implementation, considering its potential to negatively affect employee perception and organizational innovation efforts.

AI Trailblazer Takeaways: This underscores that adopting AI represents not merely a technological shift but also a dramatic change management challenge. Start small with one team, in one or two functions, versus integrating AI into your organization in a heavy-handed fashion. Prepare your organization for cultural change. This shouldn’t just be the CTO’s responsibility.

Perplexity announces a move to sell ads.

In a surprising move, Perplexity announced via an exclusive in Adweek that it is going to start selling ads on its search platform. Here are the highlights.

• Funding and Ads Introduction: Perplexity AI, a Google competitor, plans to sell ads after raising $73.6 million in Series B funding led by investors including Jeff Bezos. The platform will introduce native ads in related questions, which constitute 40% of its queries.

• AI-Driven Search Experience: Utilizing AI, Perplexity answers user questions with web sources, incorporating multimedia and data from partners like Yelp. Despite initially promoting an ad-free search vision, advertising was always planned to support its business model.

• Subscription and AI Technology: Founded by researchers from Meta and OpenAI, Perplexity combines OpenAI's GPT with its AI to offer a rich search experience. It operates on a $20 monthly subscription and plans to embed brand-sponsored questions seamlessly.

• Advertiser Appeal and Challenges: The native ad format is seen as non-disruptive, but Perplexity faces challenges in scaling its user base, ensuring brand safety, and maintaining ad relevance. Success depends on its over 10 million monthly active users and the implementation of effective targeting.

• Brand Safety and Implementation: As Perplexity moves towards advertising, brand safety and the relevance of sponsored content are key concerns. Ensuring sponsored questions align with brand values is critical, with Perplexity prioritizing brand safety in its upcoming ad product developments.

AI Trailblazer Takeaways: Originally heralded as the first real threat to Google’s dominance in a while, this move may make Perplexity look and feel like yet another Google wannabe. The move also shows that subscriptions aren’t always enough of a business model for consumer AI startups. For marketers, competition in Search is always good. Time will tell if Perplexity becomes a viable alternative or not.

What is AI Trailblazers?

AI Trailblazers is a vibrant platform dedicated to uniting marketers, technologists, entrepreneurs, and venture capitalists at the forefront of artificial intelligence (AI). We launched at CES in partnership with The New York Times and hosted our members for an evening of AI programming at SXSW. Below is how our SXSW dinner guests rated their experience.

AI Trailblazers is designed to fuel growth, innovation, and career development among its members, who all want to be at the forefront of incorporating artificial intelligence into their businesses and their lives for strategic advantage. More information here.

Links of the week

How should I be using A.I. right now? (NYT)

The most promising artificial intelligence startups of 2024 (CB Insights)

Business schools are going all in on AI (WSJ)

The Deodorant AI Spokesmodel Is a Real Person, Sort Of (New York Magazine)

Generative AI’s use isn’t meeting its perceived marketing potential—yet (EMarketer)

The 2024 MAD (Machine Learning, AI & Data) Landscape (Matt Turuck)

Adobe’s GenStudio brings brand-safe generative AI to marketers (Techcrunch)