Predicting the Future by Reflecting on the Past.
Researchers have devised a new method to coax ChatGPT into its most daunting task: predicting the future.
According to recent studies, the most effective approach for obtaining accurate predictions from ChatGPT is to prompt it to narrate a story set in the future, reflecting on events that have yet to unfold.
By evaluating 100 different prompts, researchers compared direct predictions (like guessing the Best Actor at the 2022 Oscars) with "future narratives." These narratives involved scenarios such as asking the AI to craft a tale about a family watching the 2022 Oscars on TV, describing the moment the presenter announces the Best Actor winner.
Interestingly, the narrative approach yielded more precise results. Similarly, for forecasting interest rates, the model's storytelling about Fed Chair Jerome Powell looking back on past events proved most effective. Redditors experimented with this technique, resulting in predictions like an interest rate hike in June and a financial crisis in 2030.
Theoretically, this means requesting ChatGPT to compose a news story for Cointelegraph set in 2025, reflecting on the year's significant Bitcoin price movements, could provide a more accurate forecast than a direct inquiry.
However, two potential issues arise:
The researchers chose the 2022 Oscars as a benchmark, assuming ChatGPT would be unaware of the outcomes due to its training data cutoff in September 2021. Yet, instances exist where ChatGPT seemingly possesses information beyond its training data.
There's speculation that OpenAI might have intentionally hindered ChatGPT's predictive responses, implying this technique could merely be a workaround.
These findings shed light on an innovative approach to leverage ChatGPT's capabilities in predictive storytelling, yet uncertainties remain regarding its reliability and limitations.