📄 中文摘要
在一次针对 NCAA 锦标赛的预测实验中，使用了六个来自不同公司的前沿 AI 模型，包括 Claude、GPT-4o、Gemini、Grok、Llama 和 DeepSeek。研究的核心问题是这些模型是否真的存在思维差异，还是因为训练数据相似而产生相同的输出。通过对比各模型的预测结果，发现它们在锦标赛的胜者预测上达成了一致，表明这些模型可能在某些方面受到相似数据的影响，导致输出结果趋同。
我让六个 AI 模型分析同一场 NCAA 锦标赛，他们的看法一致。
出处: I Gave 6 AI Models the Same March Madness Bracket. They All Agreed.
发布: 2026年3月19日
📄 English Summary
I Gave 6 AI Models the Same March Madness Bracket. They All Agreed.
A prediction experiment for the NCAA March Madness tournament utilized six frontier AI models from different companies, including Claude, GPT-4o, Gemini, Grok, Llama, and DeepSeek. The central question was whether these models think differently or if they produce similar outputs due to being trained on the same data. By comparing the predictions from each model, it was found that they all agreed on the tournament winner, suggesting that the models might be influenced by similar data, leading to converging outputs.