Nvidia is one of the biggest winners in the GAI brain rush. Since late May when its revenue forecast beat expectations, the chip designer has added 47% its market capitalization, which totaled nearly $1.1 trillion on July 24,
With Microsoft
MSFT
Their answers suggest GAI is on the rise — which bodes well for Microsoft’s ChatGPT.
While they’re uncertain how it will evolve, they envision an optimistic scenario in which GAI saves people time, helps boost business productivity, and possibly enables companies to increase revenue while abiding by humanistic values.
To become widely adopted, Li and Argenti agree GAI must create tangible business value, avoid violating social norms, and comply with an organization’s policies and regulations.
Argenti — who considers GAI an innovation as significant as the printing press — describes how Goldman is experimenting with many applications to find the ones most likely to add value over the long term.
Goldman’s process strikes a balance between widespread idea generation and effective investment aimed at achieving high “return on attention,” he said.
High-Value Uses Of GAI Will Succeed And Spread
Argenti rates Large Language Models among history’s most important innovations. As he told me in a July 20 interview, “From what I’ve seen, [LLMs] are the largest, most profound revolution in technology since the invention of the printing press. [LLMs] are a revolution in how we manage and deal with knowledge in a scalable way.”
LLMs break two barriers that previously separated knowledge from people. “Before the printing press, people needed to travel to be in physical proximity to books and be educated well enough to read and understand them,” Argenti said. “The printing press made it possible to copy books and spread them to libraries that were closer to people.”
While the printing press made books more physically available, they were no easier to understand — until LLMs — which “break the comprehension barrier.” As Argenti said, “They explain something difficult in simpler terms. They put the reader and the writer at the same level. This doesn’t just happen — it depends on how you ask questions. It works better if you break it down in steps.”
To succeed, GAI must pass the test of any successful business — it creates so much value for people they are willing to pay for it. As Li wrote in a July 17 email, “GAI will be judged on the same basis as any business — whether it’s contributing to a technology or product that people actually value. From the business side of it, [that means] a use that enables the company providing it to be paid in a sustainable way.”
Another important test is whether GAI applications offer unique value. “If you use (say) OpenAI tools to provide a product that anyone else with access to OpenAI tools can also build, you are really just a commodity service provider and won’t stay in business long if you don’t add something unique,” noted Li.
Moreover, GAI will not succeed if it merely makes gauzy promises not grounded in reality. As she noted, “One can imagine thinking about socially valuable things like combating climate change.” Business leaders should view appealing buzz phrases — such as “generative AI for sustainability” — with a skeptical eye to discern whether it will result in a “concrete” solution.
The most transformative GAI applications will do things that could not be done before. Many will begin their GAI journeys by speeding up something they already do — such as “coding, making small talk, drawing something cute, or brainstorming by combining old ideas.”
Yet Li sees the most potential value coming from using GAI to create “something we couldn’t have done before.” In her view, GAI “will be used and normalized.”
Goldman is focused on GAI applications that quickly enable its professionals to deliver clients the most insight. As Argenti said, “We are a digital business and aim to provide information that is pertinent for a context. Time is the most valuable currency. We aim to invest in GAI applications that provide a return on attention.”
As he explained, such applications include:
- Helping developers write code — which “lowers the barriers to entry” for them to use Java, Python, and Slang.
- Instrumenting complex operations — for example, enabling a portfolio manager to “query a basket of stocks” — applying criteria such as “technology companies that will benefit from AI, are not exposed to China, and have diverse boards.”
- Empowering brokers talk to clients about market moving events. For example, when one of these occurs, the application enables the broker to see how the event will change the value of a client’s portfolio and provide a list of talking points the broker can use in conversing with each client.
Companies Must Invest To Build Effective GAI Applications
Goldman is actively testing use cases while making sure they are consistent with its governance standards — such as accuracy and intellectual property protection. Goldman enables “safe experimentation before rolling out applications more broadly,” Argenti said.
While Goldman brainstorms hundreds of use cases, it invests resources in those that satisfy criteria such as:
- Is GAI the right technology for solving this problem?
- Does the application help augment talent?
- Will the application have a long-term impact?
- Will iterative feedback make the application better?
The economic value of a generative AI chatbot can be significant. It can make call center workers more efficient, provide a better customer experience, and spur customers to return more frequently.
As Li explained, “A reasonable short term positive outcome might be that you handle calls faster and can divert more routine calls to the bot and save the time of your best human performers for more difficult tasks. A great scenario is that your customers are so pleased with these faster better interactions that they call more often and you have a way of doing customer discovery this way.”
Such positive outcomes are more likely if a company’s culture puts a high value on happy employees and customers. The consequences of developing GAI applications with the opposite culture are predictably bad.
As Li noted, “You could train [the chatbot] poorly — then you lose customers who are frustrated or you end up facing huge liabilities. You could also adopt [GAI] in a way that alienates your best employees and leaves them looking for ways to leave. Then you have employees that want to leave and your bot is not good enough to actually replace them.”
GAI’s Benefits Will Exceed Its Costs And Risks
Li envisions the benefits of GAI could contribute to significant economic growth. Whether its benefits exceed GAI’s costs depends, she said, “on circumstance and also policy and management decisions.”
Policies can shape that outcome. As Li noted, “One can imagine various policies that both firms and regulators can take regarding how to address how AI is used as well as policies related to how profits from growth are redistributed.”
Another determinant of GAI’s ultimate success depends on how job productivity is defined. As she wrote, “Generative AI learns on examples. If I provide a good example of how to interact with a worker, and the machine learns from that, do I get paid for it? The writer’s strike in Hollywood right now relates to ‘residuals’ and I think the idea of residual value will become more important in other jobs as well.”
How Companies Can Get Value From GAI
Goldman Sachs expects GAI’s benefits to exceed its costs as long as the bank makes the right decisions. As Argenti said, “In theory, a company can spend hundreds of millions of dollars building a large language model. However, a company can get much more value by pursuing the right policies.”
These include using constitutional AI — which articulates technologies’ goals and the values that bound how those goals are pursued. A company can train an AI tool with a combination of external searches and company specific data. “You don’t need to build a fully pre-trained model— you can build your application on top. And you can engineer your prompts to gain better insights,” he explained.
Implications For Microsoft Stock
Microsoft — whose stock is up more than 43% in 2023 — is expected to post 7% revenue growth for its fiscal fourth quarter to $55.5 billion, according to Yahoo! Finance, while adjusted earnings are forecast to rise to $2.55 — up 14.3% from the year before.
Microsoft could add $14 billion a year to its revenue from CoPilot — the software giant’s ChatGPT-powered assistant for Microsoft Office products like PowerPoint, Word, and Excel. Macquarie Equity Research made this revenue estimate by assuming 10% of Microsoft’s 382 million customers will pay CoPilot’s $30 a month fee.
If Li and Argenti’s optimism about GAI is right, the payoff from Microsoft’s $10 billion investment in OpenAI could make its outlook very bright — boding well for those who own its stock.
Read the full article here
