samuel rizzo

@xxS0ares @robertotcestari quando fiz usei HMA

@rizzo_exe @robertotcestari Qual vpn vc recomenda ?

A assinatura do Google AI ultra custa 100 dolares no mundo inteiro, mas no brasil ehh 780 reais 🤡

@1kartikkabadi1 @cursor_ai hey, are they still charging per request? like, with $20 you get 500 requests? it’s been a long time since i last used cursor, but i think i’ll give it another try

Just got @cursor_ai Pro to test out Composer 2.5 and it's pretty good. Thanks for the 10x limits, time to tokenmaxx

@LexnLin now it only needs one more thing: a really good model

nvm just a copy

@DavidOndrej1 the cursor agent is still by far the best. opus 4.7 there genuinely feels like magic, the problem is the pricing.. since anthropic is basically subsidizing claude code, it becomes very hard to compete

everyone uses Cursor... lol

@TTrimoreau yes, but why code without AI these days?

Can you CODE without AI...?

@mark_k @xai bro, do you know if @xai has any plans to provide composer 2.5 in grok build?

Signs of life from Grok 5 at @xai

@isaac_ts_way i have no idea why there’s still no option to run composer 2.5 inside grok build

Composer 2.5 is so good I switched back from opencode to Cursor I don’t like the cursor ui nearly as much but composer 2.5 can seemingly do the most difficult stuff i need and the speed feels so good

@servasyy_ai we built our entire workflow around @karpathy’s methodology, and our shipping speed became insanely fast with way fewer review cycles

Andrej Karpathy：“Claude 的错误里，90% 来自上下文缺失，而不是模型能力弱。” 没有 `CLAUDE.md` 时，错误率是 41%。 有 4 条基础规则时，错误率降到 11%。 使用下面这套 12 条规则时，错误率降到 3%。 这是资深工程师最终沉淀出的 12 条规则： 1. 写代码前先思考 先说明假设，不要猜。模型不会读心，别指望它能自动知道你的意思。 2. 简单优先 最少代码，不做投机式抽象。一旦你让 Claude 为“未来灵活性”加东西，可能就多出 200 行下季度要删的代码。 3. 外科手术式修改 只改必须改的地方。别让它顺手优化旁边的代码，PR 就是这么膨胀的。 4. 目标驱动执行 先定义成功标准，然后循环直到验证通过。没有成功标准，Claude 要么无限循环，要么过早停止。 5. 只把模型用于判断型任务 比如分类、草稿、总结、抽取。不要让模型处理路由、重试、状态码处理、确定性转换。代码能回答的，就让代码回答。 6. Token 预算不是建议 单任务 4000，单会话 30000。长时间调试到第 40 条消息时，Claude 会重新建议你第 5 条消息已经否掉的修复方案。 7. 暴露冲突，不要折中平均 代码库里有两种模式？选一种。Claude 把两种混在一起，错误就会被吞两次。 8. 先读再写 先读 exports、调用方、共享工具。Claude 很可能在一个已有相同函数旁边，再加一个重复函数，只因为它没读到。 9. 测试要验证意图，而不只是行为 如果业务逻辑变了测试却不会失败，这个测试就是错的。Claude 写的 12 个测试都可能通过，即使函数实际只返回一个常量。 10. 每个重要步骤都要 checkpoint Claude 可能在第 4 步已经坏掉的状态上继续完成第 5、6 步，而没人发现，浪费一小时。 11. 匹配代码库约定 项目用 class components，就不要默默改成 hooks。测试模式可能依赖 `componentDidMount`，hooks 会破坏它，却不一定暴露问题。 12. 失败要大声暴露 “成功完成”，但 14% 的记录被静默跳过，这是最糟糕的一类 bug。要暴露不确定性，不要藏起来。 真正会复利增长的，不是下一个框架，而是： - 把 `CLAUDE.md` 当作跨会话的组织记忆 - 基于 eval 改进，而不是凭感觉改 - 重视 checkpoint，而不是一味追求速度 - 明确暴露冲突，而不是静默混合 - 纪律永远比框架重要 - 一个仓库，一个规则文件，没有例外 在这件事变成 AI Twitter 的大众共识之前，提前领先几条规则。 研究它👇

@nicdunz simple bypass: instead of saving the image, just take a screenshot

you can now check if an image was generated by openai

hey @bcherny, how can I properly report this issue to your team?

@VictorTaelin gpt 5.5 é o desenvolvedor que tem os diplomas opus 4.7 é o desenvovedor que tem as manhas

@zavxai no

Be honest, GitHub commits really matters?

@VictorTaelin muito hype pra pouco resultado

...

The Claude Code hackathon is back for Opus 4.7. Join builders from around the world for a week with the Claude Code team in the room, with a prize pool of $100K in API credits. Apply by Sunday: cerebralvalley.ai/e/built-with-4…

@loudgods entrei aqui mas apareceu o app que eu fiz, você me copiou?

claude code é simplesmente insano (Opus 4.7) eu sei LITERALMENTE nada de código. Acabei de construir um app totalmente funcional em minutos. ➡️ http://localhost:3000/

@bielconn @grok responda de forma simples e concisa

Uma pergunta: caso a NASA errasse alguns km de distância da Lua, e a nave dos astronautas fosse puxada pela gravidade da Lua, como eles escapariam da Lua se acabarem pousando sem planejamento?

@NASA @grok what is the latency from the farthest distance reached during the artemis ii mission to earth data centers?

Make new friends, but keep the old. A new photo captures the Moon's near side on the right (the side we see from Earth, identifiable by its dark splotches) and its far side on the left. The Artemis II crew are the first to see the far side with human eyes.

@meunomeebero vps + dokploy, resolve qualquer saas mano

Muito se fala de lovable,vercel... pra subir SaaS mas eu continuo achando VPS >>>> Nada como um docker compose com um bun, tanstack, postgres, nginx e um certbot Um comando pra subir dev e um comando pra subir prod, delicia demais VPS é um caminho sem volta

@forgebitz software is solved, devops definitely isn’t

software is solved can't even use claude in an app for end users, truly amazing