OpenAI's ChatGPT drew attention as it was the first AI participant in a secret prediction contest involving economists and hedge fund investors. According to a report by Business Insider on the 6th (local time), this year marks the seventh year of the annual contest, run by economist David Seif. Participants predict about 30 events in various fields, including politics, economy, and sports, and rank them by the square of the difference between the prediction and the actual result. The lower the score, the more accurate the prediction was.
Hedge fund director Sam Leffell decided to involve ChatGPT, entering complicated game rules and 30 questions. ChatGPT understood the rules and presented probabilities for each event in just a few minutes. Human participants, on the other hand, spent many hours over several days understanding and investigating the questions and calculating the probabilities.
ChatGPT ranked 80th out of 160 when the competition ended in November. It is in the middle of the list. Safet said ChatGPT performed well when there was a lot of data that was helpful for forecasting but failed to reflect the latest news. In fact, ChatGPT predicted a 95% chance of the astronaut's return, but missed the news announced at the end of 2024.
But Reppel offered a different perspective. "Chat GPT performed better than half of the people and spent much less time than everyone else," he said. "If you look at the results per hour, maybe Chat GPT won."
"What if we have to predict 30,000 events quickly instead of 30?" said Leppel, an investor. "It's become commonplace in both personal life and work. ChatGPT is now basic," he added.
■ This article is provided through a partnership with AI media outlet AI Matters. The article was written using the Claude 3.5 Sonnet and ChatGPT.
AI Editor ([email protected] )
Stock investing still seems to be at the level of a coin toss even for artificial intelligence.
With so many complex factors and reactions involved, it ultimately becomes a game of probabilities.
There are program trades, but also human decision-making, which makes analysis even more complicated.
Still, at a moderate level, it seems possible that results could emerge before long.