Tag
This paper introduces a linguistically motivated phonemic tokenizer for Vietnamese and Chinese that factorizes syllables into onset, rime, and tone components, reducing vocabulary size and improving efficiency. The proposed PhonemicBERT models demonstrate competitive or superior performance in language understanding tasks compared to existing tokenizers and pretrained models.
HSK Manga is a free tool that simplifies manga dialogue to HSK levels for Chinese learners, using AI and OCR to make reading more accessible for beginners.
The Chinese tutorial on the ICT trading model, created using Fable 5, is now online, systematically explaining the core concepts and trading system.
Introduces the Chinese tutorial *Hands-On Pi*, covering the step-by-step implementation of a Pi-style coding agent from scratch through 15 runnable checkpoints, including protocols, tool loops, conversation tree storage, context compression, and evaluation. Each chapter comes with the main text, a corresponding commit, tests, and failure experiments.
Recommend using tgzhidao.com, a Chinese Telegram navigation site. Users can search channels, groups, bots, and tool resources by keywords, avoiding the need to search for a needle in a haystack in Telegram.
Recommend a website called TG知道, which specifically collects Chinese Telegram channels, groups, and bots, supports categorized search, making it convenient for users to discover quality resources.
Introduces Hy-MultiTurn, a Chinese benchmark for deep multi-turn dialogue understanding, evaluating 22 frontier models across six controlled modes (constraint memory, precise execution, constraint synthesis, object localization, action suppression, reference resolution) with 209 tasks spanning 12-76 turns. Even the strongest model, GPT-5.5, satisfies only 41.1% of responses.
Jinchenma shares how to remove AI flavor through an iterative personal writing system, including clarifying author identity, article types, taboo rules, and methods for multiple independent reviews and rule iteration.
The mimofan tool has built-in 29 features, including tool call loop breaker, auto-compaction mechanism, token usage tracking, system prompt optimization, etc. It is a feature-rich AI agent tool.
CallBench is a Chinese benchmark for evaluating dual-goal coordination in phone call assistants, containing 50,000 multi-turn dialogues across six scenarios, with a preset-aware evaluation protocol covering semantic understanding, safety, and dialogue rhythm.
Introduces a GitHub project called TelegramGroup that curates a large collection of high-quality Telegram groups, channels, and bots, categorized by direction to help newcomers quickly find the resources they need.
Recommends an open-source Chinese AI Agent general knowledge tutorial website, including 21 chapters with visualized animated tutorials, a 130+ question bank, and an AI teaching assistant. It is purely static and can be deployed to Nginx with one click.
MedRealMM is a new multimodal benchmark for Chinese online medical consultation, built from real-world patient-doctor interactions, evaluating LLMs on next-response generation with clinical rubrics.
A curated Chinese summary of OpenAI's official GPT-5.6 prompt guide, including core changes, recommended structure, templates, and common pitfalls.
Introduces two open-source AI writing skill templates that mimic Liu Run's business analysis style and Luo Zhenyu's quote style, enabling AI to produce professional and shareable articles, lowering the barrier to writing.
The user tested the frontend capabilities of GPT 5.6-Terra, creating an interactive Guan Yu card webpage from the Three Kingdoms. The results were outstanding, particularly in image matting and depth-of-field perspective.
Xu Chonglang released a free open-source "Investment Guide for Beginners" covering basics of US stocks, options, and cryptocurrency, aiming to help Chinese investors build a complete knowledge framework, identify risks, and make independent decisions.
CC Max will continue to offer the service after delisting fable 5, at the same price. Current test error rate is about 3%, but concerned about potential issues after launch.
余温 shares a review of his three-year AI startup journey, arguing that technology is not valuable; channel and reach capabilities are what truly matter. He also recommends a website that aggregates data from all social platforms.