The hype surrounding generative AI has created an aura of excitement and urgency among evangelists. It might lead one to believe that Fortune 500 companies are quickly embracing large language models (LLMs) and transforming corporate America into one giant chatbot. However, the reality is quite different. The implementation of generative AI is moving far slower than anticipated, even with CEOs pressuring C-suite executives to develop AI-centric strategies. A KPMG study revealed that a majority (60%) of U.S. executives expect generative AI to have a significant long-term impact, but they are still at least a year or two away from implementing their first solution.
Even renowned companies like Goldman Sachs are treading slowly and carefully when it comes to generative AI. Despite nearly a year passing since the release of ChatGPT, Goldman Sachs has yet to put any generative AI use cases into production. Instead, the company is deeply engaged in experimentation and has set a high bar of expectations before deployment. As a highly-regulated entity, Goldman Sachs understands the importance of careful deployment. Marco Argenti, CIO at Goldman Sachs, emphasized the need to feel comfortable about the accuracy and correct management of information before deploying generative AI into production. The company is not randomly running AI models; it has implemented a platform that ensures technical, legal, and compliance checks.
To build confidence in generative AI, Goldman Sachs has taken several measures. The company has implemented a front-end server that filters out inappropriate content, while also logging all interactions for authorization purposes. These measures ensure a streamlined and compliant user experience. Despite these precautions, Argenti stated that the company does not plan to build its own LLM from scratch. Instead, Goldman Sachs is fine-tuning existing models and utilizing retrieval-augmented generation (RAG), an AI framework that retrieves facts from an external knowledge base to ground LLMs in accurate and up-to-date information. This approach leverages the company’s valuable data while combining it with the power of generative AI.
Like many other enterprise companies, Goldman Sachs is focused on determining the ROI of generative AI. Argenti noted that everyone is seeking confirmation of the usefulness and profitability of these investments. Goldman Sachs is ready to expand its generative AI experimentation beyond software development but in a methodical and thoughtful manner. The company aims to strike a balance between productivity enhancement and investing in technologies that can potentially reshape their business models. Argenti believes that a hyper-focus on productivity alone will not lead to differentiation, as it will eventually become a new baseline for all companies. Instead, they are exploring how generative AI can transform the roles of advisors, investors, and traders.
Goldman Sachs’ approach to generative AI is practical and grounded in specific use cases. While the company has made bold statements about AI’s potential to make workers “superhuman,” they remain cautious in their experimentation and testing. The CEO and board of the company have been supportive of their generative AI efforts, but their trajectory of adoption remains slow and deliberate. Argenti described the company’s approach as having “a lot of horses in the race.” This means that while they are not rushing into generative AI, they are also not falling behind.
The slow adoption of generative AI in enterprise companies, like Goldman Sachs, demonstrates the careful and deliberate nature of implementing such technologies. Despite the generative AI hype and pressure from CEOs, companies understand the importance of accuracy, compliance, and ROI. They are taking the time to experiment, test, and build confidence before deploying generative AI use cases into production. While the pace may seem slow compared to the hype, it ensures that the implementation is thorough, successful, and capable of reshaping business models in a sustainable manner.
Leave a Reply