GPT-4 Passes Bridgewater's Investment Associate Test

Can you use a large language model as an investment analyst?

According to one executive at Bridgewater, we could be closer than we thought to integrate AI language models into financial decision-making and investment research.

The Role of AI and LLMs at Bridgewater

Earlier this month, Bloomberg interviewed the Co-CIO of Bridgewater Greg Jensen who revealed that OpenAi’s GPT-3.5 and GPT-4 has successfully passed Bridgewater’s investment associate test.

Because all of a sudden if you have an 80th percentile investment associate, technologically, you have, you know, millions of them at once.

Jensen stated how impressed he was with ChatGPT’s abilities and investment knowledge. He also noted that the LLM requires a lot of training and providing it with a ‘rigorous statistical backdrop’.

And if you have the ability to control their hallucinations and their errors by having a rigorous statistical backdrop, you could do a tremendous amount at a rapid rate.

Jensen does not believe that ChatGPT will completely replace humans on the trading floor yet, although he does note that he is working on a fund that is completely run by ‘machine learning techniques’. 

For now, it seems that Bridgewater will continue experimenting with using GPT for trading. He refers to the LLM as having

...millions of junior staffers working all at once

which could give some insight into the scale at which Jensen is looking to use GPT for investment research. Jensen remarked on how close we are to LLMs being able to ‘mimic human intuition’ rather than just providing outputs of information. 

The Process of Theorizing and Testing with LLMs

Bridgewater is focused on providing a statistical backdrop to ChatGPT in order to provide more accurate and fulsome outputs. Like with any LLMs, it is also important to learn how to engineer the right prompts in order to get a response that has the desired accuracy and reasoning behind it.

As Jensen puts it:

"Language models can come up with potential theories...then you can use statistical things to control that...Then you can use language models again to take what's coming out of that statistical engine and talk about it with human or other machine learning agents and kind of report back on what you're finding."

From this point of view, it seems that Jensen is looking at ChatGPT as a way to collaborate with humans, rather than being the sole decision-maker.  

Summary: LLMs for investment research

Given that ChatGPT was only released 7 months ago, and GPT-4 was only released in March, it's quite impressive to see it be able to pass an investment research associate test at the world's largest hedge fund.

It's clear from these test results that LLMs will play an increasingly important role in investment research in the coming years.