与将棋 AI 互动的收获:在 Floodgate 中获得第一名的路径及我对精简模型的看法

📄 中文摘要

通过与开源将棋软件的互动,作者在 Floodgate 在线将棋服务器上取得了超过 4500 的评分,并在 200 场比赛中达到第一名的排名。这一成就是在接触开源软件后约两个月内实现的,作者自 2025 年 12 月开始编程。文章并非实施指南,而是分享了通过将棋 AI 获得的经验,以及这些经验如何应用于大语言模型(LLM)研究,尤其是在推理优化和模型选择方面的挑战。

📄 English Summary

What I Gained from Interacting with Shogi AI: The Path to 1st Place in Floodgate and My Approach to Distilled Models

Interacting with open-source shogi software led to achieving a rank of 1 on Floodgate, with a rating exceeding 4500 after playing over 200 games. This accomplishment was reached approximately two months after engaging with the software, following the author's programming journey that began in December 2025. The article does not serve as an implementation guide but shares insights gained from shogi AI and their applicability to large language model (LLM) research, particularly addressing challenges in reasoning optimization and model selection.

Powered by Cloudflare Workers + Payload CMS + Claude 3.5

数据源: OpenAI, Google AI, DeepMind, AWS ML Blog, HuggingFace 等