Executive Brief: Sept 28

Innovation Insights 

1. Larger and more instructable language models become less reliable (Nature)
The paper examines how larger and more instructable language models, despite improvements, become less reliable. It identifies three main issues: difficulty concordance, task avoidance, and prompting stability. Scaled-up models often provide seemingly plausible but wrong answers instead of safely avoiding responses to complex queries, leading to errors that human supervisors often overlook. While scaling increases performance for difficult tasks, it doesn’t consistently improve reliability for easier ones. The study calls for a new design approach, particularly for high-stakes AI applications, to ensure predictable error management.

2. Updating enterprise technology to scale to “AI Everywhere” (Bain)
To scale AI across an organization, CIOs must overhaul their technology functions, re-architecting their tech stack with an “AI everywhere” mindset. This requires upgrading operational models, upskilling teams, and ensuring that AI systems are production-ready. AI deployment demands seamless integration, robust data platforms, and modern architecture, including cloud infrastructure and API-driven systems. Equally important is adopting AI-first software development processes to manage complex models and workflows.

3. How companies use generative AI to execute with speed (Ideas Made to Matter)
Companies are leveraging generative AI combined with data analytics to accelerate product development and customer interactions. For example, Pfizer and Takeda use AI to streamline drug development, reducing time to market by improving knowledge transfer and clinical trial design. Other companies like Comcast and CMA CGM use AI to enhance customer service and dynamic pricing through faster, data-driven decision-making. Dick’s Sporting Goods and McKinsey utilize generative AI for quicker and more personalized communication. While speed is a key advantage, managing accuracy and workforce impacts remains critical for successful AI integration.

4. How CEOs are using Gen AI for strategic planning (Harvard Business Review)
Gen AI can enhance decision-making by offering fresh, divergent perspectives, as seen in the case of an agricultural company, Trident, where AI identified external issues like technological advancements that the team had missed. However, its predictions can be incomplete, such as missing crucial elements like profitability. Additionally, AI cannot predict the future reliably, as it works from historical data. Despite these limitations, gen AI offers valuable insights when used alongside human judgment, acting as an augmentation rather than a replacement for strategic thinking​.

5. The intelligence age (Sam Altman)
AI will act as personal assistants, tutors, and healthcare coordinators, contributing to shared global prosperity and transforming industries. Deep learning, which works better with more data and compute power, has been key to unlocking AI’s potential, and ongoing improvements will lead to even greater advances in science and technology. While there are risks and challenges, such as shifts in labor markets, AI offers unprecedented opportunities to solve global problems like climate change and explore new frontiers. The Intelligence Age will bring prosperity that is hard to fully comprehend now, just as past generations could not envision today’s world.