I Trained a Markdown File to Boost GPT-5.5 by 23 Points — It Shouldn't Work
中文摘要
通过训练一个1400词的Markdown文件而非微调权重，作者将GPT-5.5的性能提升了23分。
English Summary
An author improved GPT-5.5 performance by 23 points using a 1,400-token Markdown file without any fine-tuning or weight changes.
原文节选
I did not fine-tune anything. I did not touch a single weight. I ran a training loop whose only output was a 1,400-token Markdown file… Continue reading on Towards AI »