Marco Argenti of Goldman Sachs presented the big investment bank’s AI strategy as one that combines foundation models with smaller, more specialized models aimed at particular tasks.
When it comes to trading, Argenti noted that the bank had been honing machine learning models for many years, and the current LLMs didn’t yet offer better investing results than its current trading technology. Long-term, though, the ability to process far more data, and test hypotheses, could change that, he said.
But outside of just trying to find great trades, there is also a lot that can be done to improve efficiency and productivity for Goldman. “Last year we had something like a million meetings,” he said. “How many of them require a briefing? That’s two million hours that could be saved with AI.”
He also conjured a scenario of large foundational models, which are better at reasoning, and more specialized models, which are more accurate, working together on a complex question. “We send it first to a larger model, ChatGPT or Gemini, and get the essence of the question, and then break it down into tasks,” which are then performed by more specialized, fact-based models, with the results then returned to the bigger model. He referred to it as “orchestration” among models.
“Applications on top of the bigger model, it creates a flywheel effect,” he added.
CEREBRAL VALLEY NEW YORK TOOK PLACE ON JUNE 27, 2024, HOSTED BY NEWCOMER & VOLLEY
cerebralvalleysummit.com/
www.newcomer.co/
www.volleygames.com/
Videos made by www.storycastproductions.com/