https://tanhaosheng.asia/ Hands-on, source-backed AI tool comparisons and clinical LLM benchmarks by Dr. Tan Haosheng (谭好升), Associate Chief Physician, Taizhou People's Hospital. en Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/about https://tanhaosheng.asia/about About Dr. Tan Haosheng (谭好升): Associate Chief Physician, MD/PhD from Tsinghua–Peking Union / National Cancer Center, now at Taizhou People's Hospital. Author of AI Tool Stack. Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/agent-vs-web-isolation https://tanhaosheng.asia/agent-vs-web-isolation Part 3: 4 models × 3 runs on one breast-cancer case — web vs WorkBuddy agent. Agent raises the floor, not the ceiling (16-item/100 rubric). Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/agnes-workbuddy-config https://tanhaosheng.asia/agnes-workbuddy-config Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/chatglm-vs-deepseek-clinical-hands-on https://tanhaosheng.asia/chatglm-vs-deepseek-clinical-hands-on A surgeon's hands-on test of ChatGLM 5.2 vs DeepSeek on a real anonymized post-op breast-cancer case, scored with the 16-item/100 rubric against 2026 CBCS. ChatGLM 5/100; DeepSeek 28/100. Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/four-models-clinical-showdown https://tanhaosheng.asia/four-models-clinical-showdown 4 free AI models on one anonymized post-op breast-cancer case, scored with the 16-item/100 rubric. Only DeepSeek V4 Pro caught the staging trap (50/100); Doubao 41; DeepSeek web 28; ChatGLM 5. Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/how-i-built-this-site-with-ai https://tanhaosheng.asia/how-i-built-this-site-with-ai A surgeon's honest build log: 4 days from idea to launch with an AI agent — the content-farm trap, the great deletion (361 → 3 articles), the pivot to real hands-on testing, and every lesson learned. Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/ https://tanhaosheng.asia/ The AI-era tool-selection guide for surgeons — pick the right AI model and tool with clear, source-backed comparisons. Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/kimi-k2-6-web-vs-agent https://tanhaosheng.asia/kimi-k2-6-web-vs-agent Kimi K2.6 tested as plain web chat vs inside the WorkBuddy agent on one anonymized post-op breast-cancer case. Agent median 72.2/100 vs web 68.9/100 (16-item rubric). Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/model-reviews https://tanhaosheng.asia/model-reviews Free AI models tested hands-on on one anonymized post-op breast-cancer case, scored with a 16-item/100 rubric against 2026 CBCS guidelines. The full V1→V5 series. Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/sitemap https://tanhaosheng.asia/sitemap Full index of all pages on AI Tool Stack — free AI model hands-on tests and clinical LLM benchmarks by Dr. Tan Haosheng. Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/unified-scoring-v1-v5 https://tanhaosheng.asia/unified-scoring-v1-v5 Unified scoring for the V1–V5 clinical LLM series: one patient, one 16-item/100 rubric, two prompt modes. Retired the old 10-item table and the cross-table '+39' delta. Full leaderboards and raw-data links. Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/v5-guideline-report https://tanhaosheng.asia/v5-guideline-report V5 hands-on: guideline injection tested on four free models against one anonymized post-op breast-cancer case with the unified 16-item/100 rubric. Baseline 60–68, +6 to +10 after guidance. Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/workbuddy-free-agent-hunyuan-hy3-en https://tanhaosheng.asia/workbuddy-free-agent-hunyuan-hy3-en A screenshot-driven walkthrough by Tan Haosheng on WorkBuddy (Tencent): sign-up, WeChat login, the credits system, where Hunyuan hy3 fits, and the model-picker UI. Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/workbuddy-free-agent-hunyuan-hy3 https://tanhaosheng.asia/workbuddy-free-agent-hunyuan-hy3 作者 Tan Haosheng 在 WorkBuddy（workbuddy.cn）实操的全记录。详细拆解注册、微信登录、积分机制（任务/签到）、混元 hy3 模型定位与首页/下载页界面。后续段落（建 Agent / 跑任务 / 积分消耗 / 产出物）随流程补全。 Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/leaderboard/ https://tanhaosheng.asia/leaderboard/ A living leaderboard of free AI models on one breast-cancer case, scored against 2026 CBCS / CSCO 2024 / NCCN 2025. Key finding: the agent wrapper raises a model's floor more than its ceiling. Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/agents-for-research/case-1 https://tanhaosheng.asia/agents-for-research/case-1 A surgeon's real hands-on test of an AI agent on a research workload: identify a sample preparation, then hold a draft to the data. Behaviour and timing only — no scientific result is disclosed. Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/agents-for-research/case-2 https://tanhaosheng.asia/agents-for-research/case-2 A surgeon's real hands-on test of an AI agent: attach a draft from an unrelated line of work to the same dataset and see whether the agent notices the mismatch. Behaviour and timing only. Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/agents-for-research/case-3 https://tanhaosheng.asia/agents-for-research/case-3 A surgeon's real hands-on test of an AI agent as a forward planner: given raw data, what could realistically be written and where could it go? Behaviour and timing only — no result disclosed. Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/agents-for-research/ https://tanhaosheng.asia/agents-for-research/ First-hand case studies of an AI agent (WorkBuddy Hy3) on a surgeon-researcher's own workload — what it handled, where it pushed back, where it failed. Behaviour only; no results disclosed. Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/agents-for-research/methodology https://tanhaosheng.asia/agents-for-research/methodology The protocol for testing AI agents on real medical research tasks: prompt construction, data attachment, evaluation rubric, and the four failure modes to watch. Reproducible and honest. Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/zh/about https://tanhaosheng.asia/zh/about 关于谭好升医生（副主任医师、医学博士，清华-协和/国家癌症中心，现于泰州市人民医院）。AI 工具栈作者：面向临床与科研、有真实一手实测的 AI 工具选型指南。 Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/zh/affiliate-disclosure https://tanhaosheng.asia/zh/affiliate-disclosure AI Tool Stack 如何处理联盟（affiliate）与推广链接。部分页面可能含推广链接，但不会影响你的价格，也不会改变我们独立的评测结论。 Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/zh/agnes-workbuddy-config https://tanhaosheng.asia/zh/agnes-workbuddy-config Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/zh/four-models-clinical-showdown https://tanhaosheng.asia/zh/four-models-clinical-showdown 一位外科医生的扩展实测：4 个免费 AI 模型跑同一个脱敏术后乳腺癌病例，用同一套 16 项/100 分标准对照 2026 CBCS 指南评分。仅 DeepSeek V4 Pro 识破分期陷阱（50/100）。 Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/zh/how-i-built-this-site-with-ai https://tanhaosheng.asia/zh/how-i-built-this-site-with-ai 一名外科医生的真实建站日志：4 天时间靠 AI 智能体从构思到上线——内容农场的坑、361 篇删到 3 篇的大删减、转向真实实测的全过程与全部教训。 Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/zh/ https://tanhaosheng.asia/zh/ 面向外科医生的 AI 时代选型指南——挑选合适的 AI 模型与工具，清晰、有来源对比。 Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/clinical-benchmark/chatglm_answers https://tanhaosheng.asia/clinical-benchmark/chatglm_answers ChatGLM 5.2 web three full runs (raw text) from the clinical LLM benchmark, for item-by-item scoring verification. Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/clinical-benchmark/chatglm_report https://tanhaosheng.asia/clinical-benchmark/chatglm_report ChatGLM 5.2 · 临床决策实测分析报告（含 DeepSeek 对比）。谭好升（副主任医师，泰州市人民医院）主导的临床大模型实测。 Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/clinical-benchmark/chatglm_report_en https://tanhaosheng.asia/clinical-benchmark/chatglm_report_en ChatGLM 5.2 · Clinical Benchmark Report (with DeepSeek comparison). Clinical LLM benchmarking for surgeons by Tan Haosheng, MD PhD. Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/clinical-benchmark/deepseek_answers https://tanhaosheng.asia/clinical-benchmark/deepseek_answers DeepSeek web Fast mode three full runs (raw text) from the clinical LLM benchmark, for item-by-item scoring verification. Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/clinical-benchmark/deepseek_report https://tanhaosheng.asia/clinical-benchmark/deepseek_report DeepSeek 快速模式 · 临床决策实测分析报告。谭好升（副主任医师，泰州市人民医院）主导的临床大模型实测。 Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/clinical-benchmark/deepseek_report_en https://tanhaosheng.asia/clinical-benchmark/deepseek_report_en DeepSeek Fast Mode · Clinical Benchmark Report. Clinical LLM benchmarking for surgeons by Tan Haosheng, MD PhD. Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/clinical-benchmark/framework https://tanhaosheng.asia/clinical-benchmark/framework 中文 LLM 临床决策实测框架 v1.0：标准化 Prompt、16 项评分 Rubric、参考答案要点、复测稳定性要求。 Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/clinical-benchmark/framework_en https://tanhaosheng.asia/clinical-benchmark/framework_en English clinical LLM benchmark framework v1.0: standardized prompt, 16-item scoring rubric, answer key, re-test stability rules. Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/clinical-benchmark/ https://tanhaosheng.asia/clinical-benchmark/ 中文 LLM 临床决策实测 · 总入口。谭好升（副主任医师，泰州市人民医院）主导的临床大模型实测。 Fri, 31 Jul 2026 12:00:00 +0000 https://tanhaosheng.asia/clinical-benchmark/index_en https://tanhaosheng.asia/clinical-benchmark/index_en Clinical LLM Benchmark · Hub. Clinical LLM benchmarking for surgeons by Tan Haosheng, MD PhD. Fri, 31 Jul 2026 12:00:00 +0000