Sabitlenmiş Tweet
Yo this is where I'll write tech stuff (primarily software engineering). I dont have anything to write yet so Imma plug my blog lol
Why I love Rust (Oct 2021)
dolpheyn.hashnode.dev/3-why-i-love-r…
English
Dolpheyn
8.5K posts
Dolpheyn retweetledi
@meowkoteeq "I'm sorry doctor but i can't take this vaccine, the protein structures needed for its development were calculated by alphafold and that's genai :/"
English
@dolpheyn Aku rasa dia buat targetted ads ke programmer la. Aku pon nampak. Hahaha
Indonesia
Dolpheyn retweetledi
Those upcoming married bros Laylatul Qadr Duas hit different
The Spectator Index@spectatorindex
BREAKING: Gold falls to under $4500, set for its biggest weekly fall since 1983.
English
Delve helped them find a compliance hack with this one
Kimi.ai@Kimi_Moonshot
Congrats to the @cursor_ai team on the launch of Composer 2! We are proud to see Kimi-k2.5 provide the foundation. Seeing our model integrated effectively through Cursor's continued pretraining & high-compute RL training is the open model ecosystem we love to support. Note: Cursor accesses Kimi-k2.5 via @FireworksAI_HQ ' hosted RL and inference platform as part of an authorized commercial partnership.
English
Dolpheyn retweetledi
Trump: we attacked Iran for Israel
Rubio: we attacked Iran for Israel
Joe Kent: we attacked Iran for Israel. The US government is occupied by Israel
Epstein: i work(ed) for Israel
Solana influencer: that's ridiculous - look how small Israel is!
mert@mert
wild that most of the world's political discourse and conspiracies are about a small piece of land the size of new jersey
English
@kentcdodds Did you make claude launch a cursor cloud agent through cloudflare with codemode or what are we looking at here?
English
Dolpheyn retweetledi
Dolpheyn retweetledi
🚨 100M TOKEN CONTEXT WITHOUT COLLAPSE
> <9% degradation from 16K → 100M
> beats RAG + rerank + SOTA pipelines
> runs on just 2×A800 GPUs
we could be back
艾略特@elliotchen100
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
English
Dolpheyn retweetledi
For the first time since 1967.
No Eid prayer at Al-Aqsa mosque.
English
Dolpheyn retweetledi
"They turned oil into a shitcoin"
Zhao DaShuai 东北进修🇨🇳@zhao_dashuai
Oil prices surpassing $116 per barrel.
English
Dolpheyn retweetledi
Dolpheyn retweetledi
imagine pulling 3rd all nighter faking compliance
Karun Kaushik@karunkaushik_
when you step into the office and your founding AI engineer is on his 3rd all nighter this team never stops shipping
English
I wonder why big g havent scale this past singapore
Andy Fang@andyfang
Introducing Dasher Tasks Dashers can now get paid to do general tasks. We think this will be huge for building the frontier of physical intelligence. Look forward to seeing where this goes!
English