Herman

@Rightanglenews So did he actually say that or just that he's taxing higher wealth more and they're typically white

BREAKING - NYC voters are shocked as Zohran confirms he will be moving forward with his campaign promise to tax White people at higher rates to help alleviate burdens on black residents. “The wealth of a White household in the city is $200,000, while that of a black is $20,000.”

@kofiwest_gh Bro what do I do with a fireplace

How many times do I have to tell you guys to stop mounting your TV in 2026 ? ☹️

Why is chatgpt website so laggy

so what did I miss, composer 2 is actually kimi? Is cursor glass any good?

@elliotchen100 Sounds very cool, yet another path to research

论文来了。名字叫 MSA，Memory Sparse Attention。 一句话说清楚它是什么： 让大模型原生拥有超长记忆。不是外挂检索，不是暴力扩窗口，而是把「记忆」直接长进了注意力机制里，端到端训练。 过去的方案为什么不行？ RAG 的本质是「开卷考试」。模型自己不记东西，全靠现场翻笔记。翻得准不准要看检索质量，翻得快不快要看数据量。一旦信息分散在几十份文档里、需要跨文档推理，就抓瞎了。 线性注意力和 KV 缓存的本质是「压缩记忆」。记是记了，但越压越糊，长了就丢。 MSA 的思路完全不同： → 不压缩，不外挂，而是让模型学会「挑重点看」 核心是一种可扩展的稀疏注意力架构，复杂度是线性的。记忆量翻 10 倍，计算成本不会指数爆炸。 → 模型知道「这段记忆来自哪、什么时候的」 用了一种叫 document-wise RoPE 的位置编码，让模型天然理解文档边界和时间顺序。 → 碎片化的信息也能串起来推理 Memory Interleaving 机制，让模型能在散落各处的记忆片段之间做多跳推理。不是只找到一条相关记录，而是把线索串成链。 结果呢？ · 从 16K 扩到 1 亿 token，精度衰减不到 9% · 4B 参数的 MSA 模型，在长上下文 benchmark 上打赢 235B 级别的顶级 RAG 系统 · 2 张 A800 就能跑 1 亿 token 推理。这不是实验室专属，这是创业公司买得起的成本。 说白了，以前的大模型是一个极度聪明但只有金鱼记忆的天才。MSA 想做的事情是，让它真正「记住」。 我们放 github 上了，算法的同学不容易，可以点颗星星支持一下。🌟👀🙏 github.com/EverMind-AI/MSA

@thdxr maybe ai can generate fun passwords I can remember

ai password manager is that anything?

alright now I'm another model behind in my testing let's go

@robinebers I did end up cancelling recently. Let's see what their big news is

sometimes I think about leaving Cursor behind 🫠 it's still the best tool but I had so many conversations lately with people that feel the same yes, it's the best AI agent there is but the direction feels unclear, I sense a lack of vision wish I'd know what their grand plan is, or if they have one coming from an engineering and product owner background, this hits harder for me than others maybe, but I really hope they figure it out

@trashh_dev At least my posts can get interaction now

i’m so cooked

@yacinelearning thanks for the reminder legend

deep breath folks it’s going to be fine

so when minimax m2.7 in opencode go

@MiniMax_AI 👀

Introducing M2.5, an open-source frontier model designed for real-world productivity. - SOTA performance at coding (SWE-Bench Verified 80.2%), search (BrowseComp 76.3%), agentic tool-calling (BFCL 76.8%) & office work. - Optimized for efficient execution, 37% faster at complex tasks. - At $1 per hour with 100 tps, infinite scaling of long-horizon agents now economically possible MiniMax Agent: agent.minimax.io API: platform.minimax.io CodingPlan: platform.minimax.io/subscribe/codi…

stop ragebaiting me chatgpt

@bridgemindai Interesting benchmark scores, I've typically found glm better than kimi and kimi better than minimax

GLM 5 Turbo just hit BridgeBench. Ranked #18. Overall score: 80.2. UI score: 50.9. Security score: 63.5. 76.9% completion rate. 15 points behind GPT 5.4. 14 points behind Claude Opus 4.6. Fast and cheap. But the performance gap is massive. Speed means nothing if the output isn't there.

@LottoLabs only one I can run on my computer ngl

Bros it’s over for qwen 3.5 0.8b

@MartinSzerment @heynavtoor Yeah this sounds interesting but I'm trying to see how it scales

🚨 This Python tool just made vector databases optional for RAG. It's called PageIndex. It reads documents the way you do. No embeddings. No chunking. No vector database needed. Here's the problem with normal RAG: It takes your document, cuts it into tiny pieces, turns those pieces into numbers, and searches for the closest match. But closest match doesn't mean best answer. PageIndex works completely different. → It reads your full document → Builds a tree structure like a table of contents → When you ask a question, the AI walks through that tree → It thinks step by step until it finds the exact right section Same way you'd find an answer in a textbook. You don't read every page. You check the chapters, pick the right one, and go straight to the answer. That's exactly what PageIndex teaches AI to do. Here's the wildest part: It scored 98.7% accuracy on FinanceBench. That's a test where AI answers real questions from SEC filings and earnings reports. Most traditional RAG systems can't touch that number. Works with PDFs, markdown, and even raw page images without OCR. 100% Open Source. MIT License.

@tmuxvim Yeah I keep seeing this too it's so annoying

has anyone else noticed that GPT-5.4 often ends its responses with like, clickbait? it often promise to reveal "the one surprising X that will do Y" or something like that

wait I can't keep up it feels like we just got GPT 5.3

@thdxr best value sub already

we've increased opencode go's limits by 3x - still $10/month