AI Shifts from 'Who Has the Strongest Model' to 'Who Can Execute Tasks'

Goldman Sachs' latest report on AI commercialization reveals a significant shift in the industry's focus from 'who has the strongest model' to 'who can truly execute tasks'. The firm conducted an on-site investigation of Silicon Valley's AI ecosystem, visiting startups, top venture capital firms, and researchers at Stanford University and UC Berkeley.

According to the report, AI is moving beyond simply answering questions to executing tasks autonomously. This new phase means that businesses are no longer just looking for models with high capabilities but also need to consider how these models can be integrated into their production environments and handle complex business contexts.

The report highlights three key areas where AI is making significant strides: Agent, Model Competition, and World Models. Agents are becoming increasingly important in automating tasks, especially those that require decision-making within clear boundaries, verifiable results, and the ability to roll back errors. Examples of such tasks include invoice processing and data entry.

Regarding model competition, Goldman Sachs notes that the focus is shifting from 'open-source vs. closed-source' to different models serving different levels of work flows. While some companies rely heavily on cutting-edge models for core production environments, others are adopting open-source models for standardized tasks due to their lower computational costs.

The concept of World Models also emerged in the report, which involves AI understanding physical systems, causality, and dynamic interactions within real-world scenarios. This is a departure from traditional Large Language Models (LLMs) that rely heavily on internet data training. Goldman Sachs predicts a significant increase in computing power demands over the next five years, potentially growing 24 times, with major players like Microsoft, Oracle, and CoreWeave benefiting from this trend.