Business leaders have much to learn from Yoav Shoham.
Shoham’s impressive career provides a strong foundation for taking on competitors like OpenAI, Cohere, and Anthropic. He holds a Ph.D. in computer science from Yale, spent over 37 years as a professor at Stanford (now emeritus), and co-founded two companies that were acquired by Google.
Alongside Mobileye co-founder Amnon Shashua and co-CEO Ori Goshen, Shoham is the co-CEO and co-founder of AI21 Labs. Since its inception in December 2017, AI21 Labs has raised $336 million, with a valuation of $1.4 billion, as reported by TechCrunch.
AI21 Labs offers products like AI21 Studio, a platform for developers to customize text-based business applications, and Wordtune, which TechCrunch describes as “a multilingual reading and writing AI assistant.”
In a discussion with Shoham, I asked him about generative AI, a topic I’ve been researching while writing my recently published book, Brain Rush. Here are some of the key questions, Shoham’s insights, and the implications for business leaders.
Why is generative AI so poor at math?
Investment banks are reportedly aiming to replace junior analysts with AI, according to The New York Times. Analysts typically produce PowerPoint presentations that analyze potential investments.
I mentioned to Shoham that I had uploaded IBM’s annual reports into ChatGPT and then prompted the chatbot to create a PowerPoint presentation analyzing the company’s financial performance and prospects. The result was a single slide with a brief quote from the latest IBM shareholder letter.
How could such an AI chatbot replace investment banking analysts? Shoham quipped, “A calculator is really good with numbers.” He explained that there’s a need for mathematical operations, and large language models must be coded to understand and calculate numerical data.
The takeaway here is that LLMs are designed to predict the next word in a sentence, not perform quantitative analysis. For business leaders, this limitation presents an opportunity for developing models that specialize in numerical tasks.
Why are company boards so concerned about AI risks?
In researching my book, I found that companies fear being left behind if they don’t invest in generative AI, while also worrying about legal repercussions if the technology produces false information. Shoham described this situation as “boards going bipolar.”
“When we started AI21, nobody cared about AI—they were focused on cloud transformation. Now, there’s FOMO—every CEO wants to be an AI company. But there’s also backlash, partly legitimate, partly fear-mongering,” he added.
Shoham pointed out that Stanford has been tracking an AI index for decades. It surged in the 1980s but went quiet in the 1990s. “Now, AI is different. It’s statistics at scale,” he said.
For leaders, the lesson is to use the best tools available for the job. “People expected AI to excel at analytics and struggle with creativity, but it turned out to be the opposite,” Shoham concluded.
What is the killer app for generative AI?
A “killer app” is a feature or application that drives widespread adoption of a new technology by offering significant time and cost savings. For personal computers, it was the electronic spreadsheet; for the iPod, it was the iTunes store.
When asked about generative AI’s killer app, Shoham suggested that not all technologies have a single killer app. “The Internet has many. All technology is about automating what was previously not automated. Generative AI can automate tasks like copy editing and form filling,” he said.
I shared my concept of the “value pyramid,” which consists of three levels of AI use: overcoming creator’s block, boosting task productivity, and driving faster revenue growth. I believe the top level offers the most potential for customer value. Shoham agreed, stating, “I have not heard of this, but I agree.”
Businesses will benefit most by launching AI applications that target the top of this pyramid.
How does AI21 Labs compete?
Although OpenAI has raised significantly more funding, AI21 competes by offering innovative technologies like Jamba, as noted by VentureBeat. Shoham explained that AI21 provides enterprise customers with “lower total cost of ownership, real return on investment solving real problems, in an economically viable way.”
For instance, a bank uses AI21’s product in customer service to enable agents to access the right documents in real-time to solve customer problems. “We pioneered task-specific models that are more efficient,” Shoham said. “The business model will shift from charging per token to a software-as-a-service model based on customer value creation.”