The hype surrounding generative artificial intelligence (AI) has led to expectations that the Fortune 500 companies are rapidly integrating large language models (LLMs) into their operations. However, the reality is quite different. Despite the pressure from CEOs and the fear of missing out on generative AI, the adoption of these technologies is progressing at a much slower pace than anticipated. A KPMG study conducted in April revealed that a majority of U.S. executives (60%) believe that generative AI will have a significant impact in the long term but stated that they are still a couple of years away from implementing their first solution. This article explores the cautious approach taken by Goldman Sachs, a leading global investment banking, securities, and investment management firm, towards the deployment of generative AI and highlights the prevailing skepticism among enterprise companies.
Marco Argenti, the Chief Information Officer (CIO) at Goldman Sachs, shared insights into the firm’s stance on generative AI in a recent interview. Despite the release of ChatGPT, a popular generative AI tool, nearly a year ago, Goldman Sachs has not yet implemented any generative AI use cases into production. Argenti explained that the company is currently heavily focused on experimentation and has set high expectations before considering deployment. Compliance and accuracy play a crucial role in decision-making at a highly-regulated institution like Goldman Sachs. Therefore, the company’s cautious approach prioritizes these factors when evaluating generative AI for production.
To put its experimental generative AI cases into production, Goldman Sachs requires a high level of accuracy. The company needs to feel confident that the information generated by the AI models is correct and that the associated risks are well managed. Moreover, Goldman Sachs expects a clear return on investment (ROI) before deploying generative AI at scale. Argenti provided an example where the company has seen promising progress in generative AI’s potential for software development, resulting in productivity gains of 20-40% during experiments. However, expanding the use cases beyond software development remains uncertain, as the company remains focused on maintaining safe experimentation and setting realistic expectations.
While Goldman Sachs is not developing its own LLM from scratch, it is actively fine-tuning existing models and leveraging retrieval-augmented generation (RAG). RAG is an AI framework that retrieves facts from external knowledge bases, grounding LLMs with accurate and up-to-date information. Argenti emphasizes the significance of the data the company possesses, driving the combination of RAG and fine-tuning. This approach ensures that the generative AI tools used by Goldman Sachs are tailored to its specific requirements and regulatory constraints.
ROI and Productivity Enhancement
Argenti acknowledged that ROI is a primary concern for businesses exploring generative AI. Companies seek confirmation of the usefulness and tangible benefits of these investments. While Goldman Sachs intends to expand its generative AI experimentation beyond software development, it is cautious about overcommitting significant resources without sufficient evidence of profitable outcomes. Argenti warns against hyper-focusing solely on productivity enhancement, as this may not drive differentiation in the long run. He emphasizes the importance of investing in technologies that can disrupt and transform business operations, rather than merely sustaining productivity levels.
Goldman Sachs adopts a practical and balanced approach towards generative AI, emphasizing the concrete application of the technology in specific use cases. Argenti’s experience with early access to generative AI tools like ChatGPT and DALL-E revealed their potential in enterprise settings. Despite the support from the company’s CEO and board, Goldman Sachs continues to prioritize cautious experimentation and thorough testing while ensuring it doesn’t fall behind in the rapidly evolving generative AI landscape. Argenti conveys that the firm has multiple initiatives in progress and is actively participating in the race, albeit at a measured pace.
The adoption of generative AI in enterprise companies, exemplified by Goldman Sachs, is progressing slowly and deliberately. The cautious approach, driven by compliance and accuracy concerns, has resulted in a reserved deployment strategy. While Goldman Sachs recognizes the potential of generative AI, the company remains committed to testing, fine-tuning, and carefully scrutinizing the technology’s impact on business operations. The measured approach taken by one of the industry’s leading financial institutions serves as an indication that other enterprise companies are likely to follow suit, ensuring a thorough evaluation and implementation of generative AI before widespread adoption.
Leave a Reply