It’s been more than 18 months since ChatGPT’s debut sparked a rally that propelled AI chipmaker Nvidia to a $3 trillion market cap in June, yet billionaire investor Ken Griffin isn’t buying the hype around the technology—not yet at least.
Speaking to Citadel’s new class of interns at New York, the founder and CEO acknowledged artificial intelligence reached an “inflection point” with the rollout of large language models, but disputed that LLM-based tools like OpenAI’s own ChatGPT would over the next three years become more valuable than the top talent he’s recruiting straight out of universities.
“For a number of reasons, I am not convinced that these models will achieve that type of breakthrough in the near future,” Griffin said on Monday, according to comments reported by CNBC.
As evidence, the Citadel boss cited self-driving cars, which struggle when confronted with so-called edge cases that fall beyond the boundaries of statistically common traffic occurrences.
Most autonomous vehicles are trained by feeding them images labelled with data describing the environment around them: that way a computer can recognize for example whether what’s crossing the street is a small child or just a rubber ball. But there are often instances when a self-driving car may find itself confronted with a situation it has never seen before, and it behaves incorrectly since it lacks human intuition.
“Machine learning models do much better where there’s consistency,” Griffin said, justifying his skepticism.
The Citadel founder has tended to focus in the past on AI as a tool for enhancing human productivity rather than replacing it. Last month he told participants at the Milken Global Conference that while it has sparked huge investments in data center infrastructure and semiconductor chips like Nvidia’s for AI training, AI’s impact on business remains small.
That’s in part because executives haven’t understood how to most effectively harness the power of AI—they just know they need to tell investors like Griffin they do.
“I’ll ask the question ‘so tell me how you use AI at your firm today’, and you get these big smiles and you get these really enthusiastic answers,” he said, “that have almost nothing to do with AI.”
‘Ph.D.-level intelligence’
While having top human talent has allowed Citadel to run one of the most successful hedge funds on Wall Street until now, OpenAI chief technology officer Mira Murati believes that her team will have an important say when it comes to whether knowledge economy employees like Griffin’s Citadel staff can still outcompete her GenAI tools in the future.
Speaking two weeks ago to students at her alma mater, Dartmouth College, Murati said OpenAI’s LLM, a type of neural network known as a Generative Pre-trained Transformer, has rapidly matured over the space of a few short years.
Whereas GPT-3 was in her words still comparable to a toddler in terms of intelligence when it debuted in 2020, the current GPT-4 released in March 2023 is already at an advanced high school level.
“In the next couple of years, we’re looking at Ph.D.-level intelligence for specific tasks,” she said.
Given GPT-4 already passed qualifying exams for practicing numerous professions, including the bar examination, one might argue Murati is downplaying the power of her technology. It remains to be seen whether Griffin will change his mind about the value and cost of human workers when Murati rolls out GPT-5.
