Notei algumas coisas bem insanas no Grok 4.6. Parece que ele está "onisciente/onipresente". Se eles realmente estiverem testando o que estou pensando, a coisa vai ficar séria. 👀

What a time to be alive 🔥 • DeepSeek-V4-Flash-0731 • Qwen3.8-Max • MiniMax-H3 • Kimi K3 And still coming: • DeepSeek V4 Pro • GLM-5.5 The open-weight era is accelerating insanely fast.

É meus amigos, Deepseek v4 Flash está insano. China, vocês estão de parabéns.👏

Tried the same technical question with Grok and Gemini yesterday. Grok nailed it. Precise answer, current code examples, and even linked a solid repo. Gemini’s reply felt generic and leaned on tools that are already a bit behind. Still processing the fact that the company with the most data in the world is getting outpaced like this.

Designing the “thinking” state for Vulora. A little Gemini-inspired, but shaped around the aurora borealis. The name comes from Vulpes + Aurora — a fox, northern lights, and private financial intelligence. Tiny details, but they make the product feel alive. #Vulora #Gemma #Private #AI #Local #Thinking #DarkTheme

Over the past few months, I’ve been building Vulora, an experimental Flutter application exploring local-first AI for personal finance. The idea started from a simple question: what if a financial assistant could be useful without depending, by default, on sending sensitive data to public clouds? Vulora is a functional proof of concept that combines local AI, offline-first architecture, tool calling, local persistence, PT/EN internationalization, and a mobile interface designed for a more private and contextual experience. It is not a public product yet, nor a launched financial solution. It is a technical case study around mobile development, applied AI, privacy by architecture, and edge inference. I published a landing page with the concept, demo, and some of the engineering behind the project: vulora.com.br Feedback is very welcome, especially from people working with Flutter, mobile architecture, applied AI, fintech, privacy, or local inference. #Flutter #AI #MobileDevelopment #EdgeAI #LocalFirst #Fintech #SoftwareEngineering #Gemma

My recent takeaway from using AI code assistants: I’m writing less. But I’m reviewing a lot more. AI didn’t eliminate the work. It just changed the nature of it. Part of engineering is shifting from production to supervision. And that requires a different skill set.

When building a new AI product, I try to answer one simple question: "If token costs get cut in half or double tomorrow, does my architecture still make sense?" If the answer is no, I'm probably building something way too reliant on current market conditions. Resilient architectures survive changes in vendors, pricing, and models.

Everyone is looking at the possible drop in model prices. I think the interesting part is another. When competition begins to happen in the price, it usually means that the main product is approaching a commodity. If this happens with LLMs, the differential is no longer the model. It becomes the system you built around it. Cache, observability, tools, UX and integration are starting to be worth more than a few extra points in benchmark.

I'm building my first SaaS in public. Meet MetriLive: helping TikTok Live sellers figure out exactly where they're losing revenue mid-stream, moving past views and likes. Never done this before. No idea if it'll actually work. But I'm documenting everything here. #BuildInPublic

Estava namorando o curso.dev do @FilipeDeschamps é ai vem a @rocketseat com esse desconto Black Friday 😓 #sacanagem

Is Cloudflare broken? I can't log in right now. 👀 #Cloudflare