ChatGPT and other AI bots are bad stock pickers - but they can still help investors, Bridgewater co-CIO says
- It'd be an awful idea to use ChatGPT and other language bots to pick stocks, according to Bridgewater's co-CIO Greg Jensen.
- "That's hopeless – that is a hopeless path," he told Bloomberg Monday.
ChatGPT and other intelligent language tools are bad at picking stocks – but that doesn't mean there's not a role for AI in the world of investing, according to Bridgewater's co-CIO.
Greg Jensen said Monday that while large language models like OpenAI's chatbot are "awful" at reading the market, the world's largest hedge fund is using them to produce hypotheses that are then tested through rigorous statistical analysis.
"If somebody's going to use large language models to pick stocks, I think that's hopeless – that is a hopeless path," he told Bloomberg's "Odd Lots" podcast.
Bridgewater is instead "combining large language models that are bad at precision with statistical models that are good at being precise about the past but terrible about the future," he added. "Combining those together you start to build an ecosystem that I believe can achieve the types of things that… Bridgewater can do – but it can do it at so much more scale."
Jensen's comments seem to chime with what people love – and hate – about large language models like ChatGPT.
The AI-powered tools often make minor mistakes, sometimes "hallucinate" entirely new answers, and can't use up-to-date macroeconomic data when they make predictions.
But the bots are capable of rapidly producing huge amounts of text – and that's what Bridgewater plans to harness, according to Jensen.
Large language models already had similar capabilities to a top 20% human analyst but can generate strategies far more rapidly, he told "Odd Lots".
"It's 80th-percentile for a reason – there are flaws," the co-CIO said.
But "all of a sudden, if you have an 80th-percentile investment associate technologically, you have millions of them at once and the ability to control their hallucinations and errors by having a rigorous statistical backdrop, you [can] do a tremendous amount at a rapid rate," Jensen added.