Hi, I'm Garth and I just discovered Cursor. My codebase just got nervous. AMA :)

Connect Claude, Cursor & Codex to Delta Exchange. Get live crypto data, positions & orders in real time.

Post-training an open weight model is the most efficient way to build sovereign AI. We’ve seen it already with Cursor Composer. If your country is trying to build LLM from scratch, that’s wasting your tax money. LLM is all about dataset, and none of country has enough data as US/China.

Frontier AI maxis make the absurd claim that open-source AI is 6 months behind. Cursor took a 2-gen old Kimi-K2.5, applied just post-training, and built an Opus-4.7 level model at a low cost. Rio de Janeiro also post-trained a 2-gen old Qwen3.5 397b to hit an Opus-level open-source SoTA. Now, many companies are already post-training the newest Kimi-K2.7. If you can't see the progress of open source here, you're simply blind.

x.com/Kimi_Moonshot/status/2… Cursorも次のトレーニング始めてる(2.6ベースだと思う)から、当面個人ならKimiがいいかな GLM 5.2は値段だいぶ高くなったし、MiniMax M3はそもそも日本語崩壊してて使い物にならない(2.5からずっとそう)ので、Kimiが一番気に入ってる ベンチマークもちゃんと中立

A. GET /posts?page=10000 B. GET /posts?cursor=eyJpZCI6... One of these works fine at 1 million rows. The other still works at 1 billion rows. Which one are you choosing for scale ?

Nếu là thật thì ngon quá nhỉ? Bỏ ra 1.700 USD để tiết kiệm 440 USD mỗi tháng CEO AMD Lisa Su vừa đem lên sân khấu một chiếc mini PC nhỏ bằng quyển sách nhưng được cho là có thể chạy model 235B, ngang tầm những model mạnh nhất mà user ChatGPT Pro hay Claude Max đang phải trả hàng trăm USD mỗi tháng để truy cập. Chiếc GMKtec EVO-X2 sử dụng Ryzen AI Max 395 với 128 GB unified memory. PC này chạy Qwen3-235B, DeepSeek V3 và Llama 3.3 70B mà không cần GPU rời. AMD thậm chí còn tuyên bố hiệu năng DeepSeek R1 nhanh hơn hơn 3 lần RTX 5080. Chỉ cần cài Linux, Ollama và Open WebUI là có thể một hệ thống AI chạy hoàn toàn cục bộ với trải nghiệm gần giống ChatGPT. Theo lời shill của một ông trên X tên là starmexxx thì nếu đang trả 200 USD cho Claude Code Max, 200 USD cho ChatGPT Pro cùng Gemini và Cursor, tổng chi phí có thể lên tới khoảng 440 USD mỗi tháng. Với mức giá khoảng 1.700-2.000 USD và tiền điện khoảng 9 USD mỗi tháng, chiếc máy này được cho là có thể hoàn vốn chỉ sau chưa đầy một năm. ----- Nhưng mà dễ gì có kèo ngon thế, đây là 1 số lưu ý cần biết sau khi nghe quảng cáo Thứ nhất, "chạy được model 235B" không đồng nghĩa với việc đạt trải nghiệm tương đương các model frontier của OpenAI hay Anthropic. Kích thước model và chất lượng model là hai chuyện khác nhau. Có những model tham số ít hơn nhưng vẫn sẽ thông minh hơn ở 1 mảng cố định, chưa kể mức độ thông minh của model thương mại thường đi trước model mã nguồn mở 4 tháng. Nếu dùng model mã nguồn mở để tiết kiệm 1 ít tiền mà bỏ qua cơ hội để tạo ra sản phẩm tốt với model thông minh hơn thì không đáng để đánh đổi. Thứ hai, khoản chi 440 USD mỗi tháng thực chất là tiền thuê dịch vụ cloud, bao gồm hạ tầng, cập nhật model liên tục, tốc độ xử lý và quyền truy cập những model tốt nhất. Mua một chiếc máy local nghĩa là đổi chi phí thuê bao thành chi phí phần cứng, điện năng và công sức vận hành. Mà thật ra phần lớn anh em Việt Nam chắc sẽ không chi đến 440 USD/tháng cho AI trừ mấy game/video studio. Tức là nếu mua PC AMD về mà không biết tối ưu để bào nó tận xương thì chi phí hoàn vốn sẽ cao hơn. Nhưng hướng đi này vẫn sẽ là tương lai, các phần cứng dần sẽ tối ưu để chạy AI local, còn AI Model sẽ tối ưu để chạy ở những thiết bị bình thường nhất như điện thoại chứ không chỉ Server lớn. Thị trường mà không có cạnh tranh thì anh em user chắc bị hút cạn tiền, đã có Claude thì phải có DeepSeek đè giá, đã có NVIDIA thì phải có AMD cạnh tranh 😁

📚Kindle本「Agent Skills実践ガイド」 現役CTO @f_uto が教える、AIエージェントに専門知識を渡す技術を完全解説！ Claude Code・Cursor・GitHub Copilot対応で、復活後すぐに実践できる💪 今のうちにマスターしておけば、次は本当の使いこなしができる🔥 amzn.asia/d/0984jc7t #AgentSkills #生成AI #開発効率化

Claude Fable 5、使えた時の感動からの突然の終了...この落差、分かりすぎる😭 でも、次に復活した時こそ本気で使いこなしたくない？ そのカギは「Agent Skills」なんです💡 40以上のツールでClaude・Cursor・GitHub Copilotを自在に操る技術。復活までの今こそ、準備のチャンス✨ #AgentSkills #Claude #AI開発

まぁですよね 来週以降6倍速モード出るので、それ待つのがいいと思う Composerもどっちも使ってるけど、Cursorのエコシステムがいらないならあんまりいらないと思う Claude Codeからの乗り換えはすごい親和性高いのでそっちからならあり

I saw in another thread how Kimi2.7 outperforms opus4.8 😂 Given how many people still use GitHub copilot/cursor - there’s so big TAM of people who will be any marketing bs.

everyone's talking about giving sarvam $1B. i get the sentiment. but this is the same mistake we made with adani-ambani - pick one winner, kill everything else. paytm invented mobile payments. upi made it a public good. then phonePe, bhim came in. competition happened. then actually something moved. what if we had just given paytm $1B instead of building upi? and honestly - benchmark gemini vs sarvam on indic tasks. gemini wins. consistently. not my opinion, just data. sarvam built something real in 3 years with almost no compute. genuine respect. but respect and national monopoly are different things. india doesn't need 1 sarvam. it needs 10. the talent exists. from vaswani to rohit patil - people are there, some even want to come back. it's not a capital problem. it's an ambition problem - we're building for domestic market when we should be building for the world. cursor fine-tuned a chinese model. they'd use an indian one too - if it was good enough. bharat didn't become "sone ki chidiya" by looking inward.

FAILURE OF IMAGINATION—Sure, everybody is clearly racing for recursive superintelligence right now. And the path forward seems so obvious: just automate software engineering, ML engineering and research, and hook it back up to itself. But if automating coding was so obvious, why didn't people go at it from the get-go? If you go and read the model card from Claude 2, for example, it's mostly about chat assistant stuff, like helpfulness, red-teaming, translation, etc. The word "coding" appears just four times, while "translation" appears twelve times! Claude 3 was more about multimodality, long-context document processing, Q&A, writing, etc. It's only around Claude Sonnet 3.5 on June 20, 2024 that model card focus shifts to agentic coding. It's not as if coding was some niche use-case: GitHub Copilot came out in 2021. Indeed, OpenAI's first product named "Codex" was a GPT-3 finetuned for code. Ok, maybe the objection is that everybody knew that automating coding was going to be a big deal, but that the models just weren't good enough, or that we really needed RLVR to make it work. Cursor, which was the fastest product to $100M ARR just two years ago, had been around for years and didn't go vertical until Sonnet 3.5 came out. But then why did the labs spend so much time on a bunch of different side projects that did not help them get to automating coding? If you go back in time to 2024 and tell researchers, by the way, by the end of 2026 you won't be coding anymore, just texting the chatbot on your phone, oh, and agentic coding will be $50! billion! of ARR (remember that anthropic's valuation was under $20B in 2024), do you really think they would spend any time working on: voice models; video models; deep research; browsers (ChatGPT Atlas released seven months ago and you have already forgotten about it). The term "AGI-pilled" comes up a lot. Do you really believe in AGI, do you really understand AGI, and so on. But even the people who are the most AGI pilled at any given time do not fully grasp the full potential of the technology or where it should head; it is largely by stumbling and not by planning that prospecting for gold has succeeded. You should really just think of AGI-pilled as believing that there is a really big "there" there, to the consternation of everybody else, but even AGI's biggest believers continuing to underestimate just how big it really is. Because the broad contours of the most audacious beliefs and predictions of the AGI-pilled have come to pass (we await now the IPOs of two trillion-dollar companies, do we not?) people tend to over-estimate the certainty and accuracy of predictions from back then. But much of what has unfolded was not obvious in foresight. Again, if it was, then some very smart, highly motivated people would have made different decisions. Look, at some point in time, there were two, maybe three people on the planet who believed in scaling: Ilya, Dario, a few others. Not even Sam, who has bought more compute at this point than the GDP of medium-sized countries, believed in scaling back then. Not even Alec, who did the first GPT paper! That was a long time ago. Then more and more people began to believe in scaling. People used "-pilled" as a suffix for scaling then too. I think the difference between really being RSI-pilled and scaling-pilled is that we are now in the regime where

this is what happens when ai agents start handling interfaces instead of just data. cursor-based tools are next to disappear as more functions get embedded within navigation menus

I gave it the cursor thermal nuclear code review skill and it found so many issues with GPT 5.5's work 🤣

Cursor

Cursor

OptimAI Search MCP brings live context to the agents people already use. Claude. Cursor. Copilot. Codex. One integration layer. Everywhere you build. Under the hood, OptimAI can process hundreds to thousands of real-time signals in parallel across the public web, social platforms, and onchain data - all powered by a decentralized network. That is the difference: not another search tool, but infrastructure for agent-native intelligence. Build with it. 💡search.optimai.network -- OptimAI: the #1 AI on BNB Chain DappBay.