AI still needs to up its March Madness game
Add Axios as your preferred source to
see more of our stories on Google.
/2025/04/05/1743880277860.gif?w=3840)
Illustration: Brendan Lynch/Axios
The AI chatbots we invited into our long-running Axios AI+ women's bracket challenge this year were no match for most of the competitors among Axios readers and staff.
State of play: Not only did the AI chatbots have trouble filling out their March Madness brackets, they ended up finishing below most of the human pickers, at least for the women's tournament, which wrapped up Sunday.
Why it matters: AI bots are likely to become key helpers in every realm of human activity — fantasy sports included — but all that computing power doesn't guarantee better results, this year's experience shows.
Zoom in: Human entries led throughout the tournament, but ChatGPT had a chance to rise toward the higher end if South Carolina prevailed against UConn.
- With UConn's 23-point win Sunday, however, ChatGPT fell to 30th place of the 46 completed brackets.
- Anthropic's entry, which had UCLA winning it all (they lost in the Final Four round), nonetheless beat out OpenAI, finishing in 24th place.
Another AI entrant from 4C Predictions (entered directly by the company) also had South Carolina winning it all.
- However, with better predictions in the early round it finished tops among the bots, landing in 18th place.
Yes, but: My bracket was even worse! I finished in 32nd place, below both ChatGPT and Anthropic.
- And some AI systems, including the ones from 4C Predictions, did quite well on the men's side.
What they're saying: 4C Predictions CEO Alan Levy noted earlier this week that, while its engine didn't get the women's Final Four exactly right, it did correctly predict the teams in the men's Final Four (all No. 1 seeds).
- "This year's results highlight a growing trend: machine learning is consistently outperforming human intuition," Levy said in a statement to Axios. "While it's true that the Final Four teams were among the favorites, the true proof of AI's superiority lies in the overall bracket accuracy, which hit 80%."
- Levy says that AI's success in the early rounds "shows that AI isn't just guessing favorites — it's identifying patterns and probabilities that human intuition tends to overlook."
Between the lines: Perhaps most important, Levy says, AI makes its choices without getting emotional.
- For example, Levy said, his firm's AI picked Houston over heavily favored Duke in the men's tournament — and Houston beat Duke in an epic comeback Sunday.
- "This wasn't based on hype, fan sentiment, or past glory — it was a cold, data-driven decision that removed human bias," he said.
