 

热榜站上线历史追踪功能：1.5亿数据的高效相似性检索

2025-12-21 分类：前沿哨所阅读(2) 评论(0) 赞(0)

智谱 GLM，支持多语言、多任务推理。从写作到代码生成，从搜索到知识问答，AI 生产力的中国解法。

热榜站近日上线了历史追踪功能demo版本，该功能允许用户点击热点旁边的趋势按钮，查看最近相似的所有热点。技术实现上，系统仅存储10天的历史数据，每天约处理1500万条，总计最多存储1.5亿条数据。数据以parquet格式落地到磁盘，查询采用duckdb引擎，热点间的相似性判断通过汉明距离计算，而非AI方法，主要是为了避免AI带来的高计算消耗。开发者在分享中坦诚表示，这种基于汉明距离的方法可能存在误判情况。热榜站(tgmeng.com)现邀请用户测试并提供反馈，以进一步优化这一功能。这一实现展示了在不使用AI的情况下，如何通过传统算法实现高效的数据相似性匹配，对关注大数据处理和相似性匹配技术的读者具有参考价值。

原文链接：V2EX 分享发现

赞(0)

未经允许不得转载：Toy Tech Blog » 热榜站上线历史追踪功能：1.5亿数据的高效相似性检索

分享到

历史追踪热榜站相似性匹配

评论抢沙发

快讯

Search Function Fails After Claude Code Integration with Gemini

Users directly integrated Claude Code with Google's Gemini CLI model through the settings.json file using the Claude format address, which runs normally in practice. However, when calling the web_search tool, the system returns zero search results, raising questions: Is it necessary to rely on the CCR tool to enable internet connectivity? Or could Google's internet mechanism fail to work in non-official CLI environments even when adapted through CCR? This issue reveals compatibility challenges in AI model integration, especially conflicts between third-party tools and native search functions, providing practical reference value for developers and AI enthusiasts, helping to understand the boundaries and optimization directions of model configuration.

Original link:Linux.do

55分钟前
Claude Connection Fix: CCH+NEWAPI+CCR Technical Implementation

Users experiencing frequent disconnections with paid Claude services, leading to poor experience. By building a custom CCH circuit breaker switching system, configured with different channel providers' main paid versions. To solve continuous disconnection issues, integrated free public API endpoints, but managing multiple keys and models was cumbersome, therefore built a NEWAPI endpoint to forward public APIs to CCH. Since public endpoints don't support Claude Code and required Codex service, added CCR to connect with chat/completions endpoint. This solution demonstrates how to optimize AI service stability through technical means, providing practical reference for similar issues, showcasing the value of DIY technical sharing.

Original link:Linux.do

55分钟前
YesCode Gateway Officially Launches with Support for Claude and Multiple Model APIs

YesCode Gateway has officially launched, providing users with Claude Code API gateway services. This service supports multiple AI model backends, including Claude, Kimi, DeepSeek, and Gemini, with billing consistent with official pricing and real-time updates on token consumption and quotas. Users can subscribe to packages or purchase API quotas separately, with initial benefits available upon official website registration. Cache read/write operations and billing display have been optimized to ensure a smooth user experience. The service is based on the $200 MAX Plan account, delivering high-quality output. Ideal for AI developers and researchers to enhance work efficiency. Join our Discord community to discuss usage tips and provide feedback.

Original Link:Linux.do

55分钟前
MiniMax IPO Filing Deep Dive: Financials and Future of the AI Giant

As a leading Chinese AI company going public in Hong Kong, MiniMax's prospectus reveals revenue of $53.43 million for the first 9 months of 2025, up 174% year-over-year. The ToC business gross margin turned positive to 4.7%, while the ToB business gross margin reached an impressive 69.4%. R&D expenses are more than triple the revenue, primarily used for model training. The company maintains healthy cash flow with $1 billion in cash reserves, sufficient to support operations for 3.58 years. With 212 million cumulative users, 27.6 million MAU, and significant growth in paying users, MiniMax focuses its business model on ToC emotional companionship, AI agents, and video generation. Compared to Zhipu AI, MiniMax demonstrates a healthier business logic. Despite facing a Disney copyright infringement lawsuit, the company's development prospects remain optimistic, making it worthy of attention from tech readers.

Original Link:Linux.do

55分钟前
Analysis of Codex Long-Running Error Issues

Users running the gpt-5.2-xhigh model with codex-cli 0.77.0 on Ubuntu 22.04 frequently encounter errors after running for more than an hour. Despite having available context memory (272K with 62% remaining), the issue occurs repeatedly. Users are asking if there are specific techniques or prompts that can resolve this issue to improve stability during long-running operations. This involves performance optimization of AI models in practical applications and serves as a valuable reference for developers and AI users. At the forefront of technology, such issues can affect the efficiency of AI applications in fields like autonomous driving and chip design, making it worthy of in-depth discussion.

Original Link:Linux.do

55分钟前
AI结构化输出API暗藏隐患：响应质量下降引发数据错误风险

本文揭示了AI大模型结构化输出API的一个严重问题——尽管看似提供了更规范的数据格式，却可能导致响应质量显著下降。作者通过实际测试发现，使用OpenAI最新发布的GPT-5.2模型进行收据数据提取时，结构化输出API会将香蕉数量错误地显示为1.0，而普通文本输出API则能正确返回0.46。文章指出，结构化输出不仅更容易在简单数据提取中出现错误，还可能导致错误建模不正确、难以应用思维链推理等技术，甚至增加提示注入攻击导致客户数据泄露的风险。作者提供了完整的代码示例，让开发者能够自行验证这一现象。这一发现对AI开发者和企业用户具有重要警示意义，提醒我们在追求数据结构化的同时，不应忽视响应质量和安全性问题。

原文链接：Hacker News

56分钟前