Wall Street nodded? Goldman Sachs is also interested in DeepSeek

The open-source AI model DeepSeek has stirred waves, and financial giant Goldman Sachs is also taking notice.

According to The Information, Goldman Sachs Chief Information Officer Marco Argenti revealed this week that the company is also very interested in the DeepSeek model, but needs to conduct a security review of the model.

The review focuses on preventing security risks such as hidden data or "prompt injection." Argenti emphasized that these security issues are crucial for any company using AI model-driven applications, regardless of where the model comes from.

Analysis indicates that Goldman Sachs' interest in DeepSeek shows that open-source AI technology is increasingly recognized by large companies. Previously, open-source models lagged behind their "closed-source" competitors, but they are finally starting to catch up.

Large companies like Goldman Sachs have significant influence in technology procurement and application, and their actions often drive other businesses to follow suit. This means that the application of open-source AI technology in the commercial field may become more prevalent in the future, becoming a business trend.

Goldman Sachs: Not Engaging in Base Model Business

Like many other companies, Goldman Sachs has improved productivity through coding software. Argenti revealed that since introducing Microsoft's coding assistant GitHub Copilot to over 10,000 software developers last June, developers' coding efficiency has increased by an average of 10%. This is equivalent to adding the productivity of 1,000 software engineers to the company.

Argenti stated that AI has brought fundamental productivity improvements to investment banking, which allows banks the option to hire fewer engineers. AI is "unlocking the capacity of human engineers."

In addition to coding autocomplete tools like GitHub Copilot, Goldman Sachs is also testing more complex AI coding tools. Argenti mentioned that the purpose of these tools is to further enhance the efficiency of software development, but specific details have not been disclosed.

Moreover, Goldman Sachs is encouraging its bankers, traders, and other employees to use AI. This month, the bank launched an AI chatbot for employees to generate client presentation slides or extract insights from lengthy documents. Employees can choose AI models from Meta (Llama), Google, and OpenAI, as each model may have its advantages in handling different tasks.

Despite Goldman Sachs' enthusiasm for AI applications, Argenti made it clear that Goldman Sachs does not plan to invest in training base models from scratch, nor does it plan to release any models. He emphasized:

"We do not engage in base model business."

This is different from the approach of some fintech companies, such as the recently released DeepSeek model by Huanfang Quantitative, which demonstrates the influence of financial companies in the AI field.

Although some AI enthusiasts in Silicon Valley, like Jonathan Siddharth from Turing, believe that models may become smart enough to reason which assets will yield the best returns, Argenti remains cautious about this Argenti emphasized that AI enhances productivity but will not completely replace humans:

“In important decision-making, human judgment remains crucial. The idea of an AI Chief Investment Officer is a 'utopian' concept because these models are usually only as good as your data; they do not have magical properties.”