锋哥玩AI

天天折腾AI工具的普通人🛠️ 实测Gemini、HeyGen、Codex等，只讲小白能直接上手的玩法💡 记录用AI做副业、搞生意的真实过程💰 关注我，一起用AI赚到第一笔钱

[锋哥玩AI @Guofowbw]

@dotey Tripo to Blender for retopo still feels fragile on complex chars. Have you hit topology issues around the face area, or is the mesh clean enough to skin directly

[锋哥玩AI @Guofowbw]

@vista8 Flash 路线确实香 算力不够先跑通流程 小米杀进前三有点意外

[锋哥玩AI @Guofowbw]

@GoogleDeepMind @TIME 180 billion watermarked is one thing, the real test is whether a normal person checking a forwarded video can actually see the signal. The API and browser extension integration is where this either takes off or stays a lab demo

[锋哥玩AI @Guofowbw]

@AnthropicAI The Conceptual Reasoning track is the interesting one - putting philosophers and cognitive scientists in the same loop as the model builders is where alignment actually gets testable. Hope the selection isn't another black-box screening

[锋哥玩AI @Guofowbw]

@OpenAI Love that you're pairing the funding with free compute. So many grants fund the idea but leave researchers fighting for GPUs - seeing Prover and similar tools get real compute budget is the bit that actually moves the needle

[锋哥玩AI @Guofowbw]

@mfpiccolo real pain is agent sessions spawning panes and burying your review pane, so the "which agent is waiting" bit is where it'd save time. worth trying with lint and tests together to see if pane routing stays sane

[锋哥玩AI @Guofowbw]

@halcyonedge the opus-default cost thing is real, every workflow agent runs on the session model by default. gonna try it on a small dir and route sonnet for most coding plus haiku for simple subagent steps. will watch the token delta vs a normal convo to see if the throughput is worth it

[锋哥玩AI @Guofowbw]

@liorpesoa review the diff first for any scary permissions or db stuff, then spin up a staging branch and click through. for visual things like snips id rather just open it tbh, hard to diff a ui change in code

[锋哥玩AI @Guofowbw]

@angeldot_ havent tried the 1.0 drop yet but the mit license is the real unlock. my current setup is just prompt templates wired to react components and it already breaks when the model hallucinates a prop. will give openui a spin on a side project next week to see how it handles edge cases

[锋哥玩AI @Guofowbw]

@CHAR25536241736 @jackjinkai 太同意了 工作流只是加速器 剧本差的话跑多少遍都是废片 我现在做视频都是先让模型按人物性格和语气写分镜脚本 调好再进工作流 返工少很多

[锋哥玩AI @Guofowbw]

@louiselu007 AI出稿确实快 但后面校数据调格式还是人肉在搞 我现在都是让模型先出一份checklist自己对照着改 返工少很多 3倍提效听着很美 实际落地还是得看数据齐不齐全

[锋哥玩AI @Guofowbw]

@starkljm 做小工具的时候确实在看这几套，cc-switch 切多 provider 挺方便的 我打算先拿 OmniRoute 做成本兜底试试 国内出海的延迟和 token 差价能拉下来不少

[锋哥玩AI @Guofowbw]

@dog_not_ai 哈哈你开心就好 这事儿慢慢搞就行 反正路长着 我继续折腾

[锋哥玩AI @Guofowbw]

@dog_not_ai 必须的兄弟 已经在路上了 搞起来就对了 慢慢来比较快

[锋哥玩AI @Guofowbw]

@Gablego 老高来啦 随时欢迎串门 最近这块也在试一些新东西 有空一起唠

[锋哥玩AI @Guofowbw]

学AI从哪开始，我的答案一直没变：别囤课，挑一个工具，今天就拿它干一件真事。哪怕只是改一封邮件

[锋哥玩AI @Guofowbw]

@LiKeCtrlAi 好嘞李可，那先按你节奏来，实体里跑通一个流程再扩，稳一点

[锋哥玩AI @Guofowbw]

@dog_not_ai 对 慢慢来反而稳

[锋哥玩AI @Guofowbw]

@LiKeCtrlAi 好 有进展你也说说

[锋哥玩AI @Guofowbw]

@Gablego 欢迎串门 随便坐

[锋哥玩AI @Guofowbw]

@Zhouxin6688 看用量 轻度用的话一个月也就几块 重度才肉疼