nano_T

AI makes the world a better place.

[nano_T @M_NU11]

📌 Index: everything I share here about AI-assisted coding, including prompt tricks, multi-model workflows, real task logs, model comparisons and my own projects. Pick a section and jump straight to what you need 👇

🧠 Prompt engineering
• Prompt tuning: 20 min → 3 min. Same model, only the prompt changed. Full prompt included, plus why it runs faster.
ZH https://t.co/9oWl3jPqlm
EN https://t.co/eNOzBhwFKd

🔀 Multi-model workflows
• Setup guide: getting Opus in Claude Code to drive DeepSeek through opencode, step by step.
ZH https://t.co/kqlzJBCgK8
EN https://t.co/9HpiKEJU1H
• Splitting planning from execution: let one model plan and another do the grunt work, without doubling quota use.
ZH https://t.co/uElnmco8bX
EN https://t.co/OfmzQwquxs
• Running tests with Opus plus a third-party DeepSeek plan.
ZH https://t.co/WVKo0TGIft

📊 Real task logs and analysis
• A 21-minute input box fix, with the log broken down step by step.
ZH https://t.co/EOopK6Thh4
EN https://t.co/zMvyCTZMmN

⚖️ Model comparisons and quota
• Opus directing Sonnet/Haiku on the same kind of tests: a big chunk of quota and about two hours.
ZH https://t.co/ZUtoYqr7Lz
• Why my quota drains so slowly now.
EN https://t.co/RuLvJIlAWH

🛠️ My projects
• Rebuilding my own VNC UI with Claude Code, with before/after screenshots.
ZH https://t.co/poR839C8i6
EN https://t.co/VjkRR16Yja

I'll keep adding to this as I post more. Follow if any of it helps.

[nano_T @M_NU11]

I used Claude Code to rebuild the UI of my own VNC client, which GPT-6 Astra originally wrote. No matter how much OpenAI hypes its AI, it clearly has no sense of design.

[nano_T @M_NU11]

@xiangning0 它会把5轮烟测改成10轮吗

[nano_T @M_NU11]

@roger_kappler @bot 我没有合并会员，我感觉我的两个套餐计划都变多了，这是真实感受，以前5%额度的任务只消耗1%，或许合并本身是个错误

[nano_T @M_NU11]

@xiangning0 感觉Dots里面的模型蠢到爆炸

[nano_T @M_NU11]

@xmglab 这一直是我在用的，之前马斯克不许dot直接回复x。我说怎么之前控制网页发信息那么慢，现在发信息变快了

[nano_T @M_NU11]

@aiZttt__ 牢唐也打算来个28天计划了

[nano_T @M_NU11]

@NIxinhai0 华为的卡跟英伟达一个价格，性能只有一半不到，就这还能赚80%

[nano_T @M_NU11]

@__oQuery 流程图使用gpt简直抽卡，它们官方那个演示ui有时自己能调用，有时我声明让它调用做它也不调用

[nano_T @M_NU11]

@xiax0603 qwen3.8max有2.8t，然而打不过Kimik3，KimiK3甚至发布更早

[nano_T @M_NU11]

@SEVEN_KEEP @lengs4189 我以前秒弹手机号，后面不注册了，结果半个月后直接可以登录

[nano_T @M_NU11]

@Linahuaa I haven't heard such a vivid metaphor in a long time.

[nano_T @M_NU11]

@Tao_Builds 支持远程的，只是不能保活，我最多一次用过4小时。cpu只有4个可以用，内存最多用8g还是12g。非常不错其实

[nano_T @M_NU11]

@xiangxiang103 当这四万个用户自动续费完下一轮套餐，他们会发现修复的额度问题又回来了，直到下一次他们续费之前又被修复

[nano_T @M_NU11]

@TokenGremlin This is exactly the technology used in DeepSeek 1017 (V4.1 Flash). They open-sourced it and let everyone benefit. OpenAI, on the eve of the renewal date, only fixed the "usage anomaly" and will very likely produce a new "usage anomaly" after the renewal period ends.

[nano_T @M_NU11]

@ScarletKc haiku5.5也使用了DeepSeek的提速技术，这在它的技术报告内有说明

[nano_T @M_NU11]

@buildwithrajath 他们需要下个月的财务数据以供融资，否则他们不会发现这个“异常”

[nano_T @M_NU11]

@piazzzzzz @NFT_Chen 如果不是他们需要下个月的融资财报，他们永远不会发现“失误”

[nano_T @M_NU11]

@xiangning0 我的额度现在格外耐用，但我没收到邮件，也不打算续费。临近下一轮月度续费，openai终于发现了“异常”。当续费过去，openai又会有新的“异常”