Rahul Mistry

576 posts

Rahul Mistry

Rahul Mistry

@_RahulMistry

https://t.co/zKrWlxd2wr | https://t.co/b6JNwedy9N | https://t.co/lGYtZI6Rdi | I am just a tech | Open Source | Cracked Developer

Vadodara, India Katılım Aralık 2020
167 Takip Edilen67 Takipçiler
Rahul Mistry
Rahul Mistry@_RahulMistry·
opencode doesnt have multi model orchastration (just like cursor selecting mutliple models with same prompt) not sure what's the team's opinion on this. @thdxr @adamdotdev
English
0
0
0
31
Charlie Holtz
Charlie Holtz@charlieholtz·
You can now quick start a repo with gstack initialized! Conductor will automatically: - setup gstack, install skills - create a new repo - kick off /office-hours
English
27
9
230
38.7K
Rahul Mistry
Rahul Mistry@_RahulMistry·
@charlieholtz dont tell me next experiment mode: adding other models soon like kimi k 😄
English
1
0
1
124
Charlie Holtz
Charlie Holtz@charlieholtz·
Codex in Conductor now has plan mode, fast mode, and skills!
Charlie Holtz tweet media
English
12
3
230
11.9K
Tanner Linsley
Tanner Linsley@tannerlinsley·
Ghostty was fun, but time for something else. I still love opencode, too but with CC plans dead on it… I’m feeling lost. Full GUI? T3 Code? Opencode GUI? Warp? Back to cursor? Try CC again? Raw Codex? My 🧠 hurts and I just need to keep shipping.
English
412
9
1.2K
325.2K
Charlie Holtz
Charlie Holtz@charlieholtz·
In general I think Conductor has a lot of room for improvement but we put a lot of thought + care into the UI. With real human brains. We try and iterate extremely fast. A few of the things you pointed out are already fixed in our latest release. 1. this is fixed in the new sidebar (settings → experimental). I agree it's confusing. the reason it’s like this is the worktree name and branch name can diverge. 2. I like our loading indicator :) 3. Tab bar needs a loading indicator too if you have multiple tabs. Would be weird if it didn’t have a loading indicator sometimes 4. True we probably don’t need this one 5. This one shows in plain English what the workspace is doing. Working/Ready to merge/PR link etc. > why workspace, these are repositories. True-ish, it’s a list of workspace grouped by repositories. > Both Does same thing One creates a new workspace in the repo, another lets you pick across repos. But agreed it’s a bit unnecessary > why is this down? This should replace the top + icon. I don’t think it should replace it, because the top + icon is a “new workspace” button, not a new repository button. but some users have gotten confused by this so we added a add repository section to the top one button too (update to latest release)
English
7
0
53
2.5K
Parth Jadhav
Parth Jadhav@ParthJadhav8·
ONG Conductor feels such a vibe coded app. - Almost nothing does what you expect it to do. - Everything feels like, it exists because AI added it - No Afterthought to any design decision ? Just pointed out some mistakes :)
Parth Jadhav tweet media
English
9
1
17
3.8K
Rahul Mistry retweetledi
can
can@marmaduke091·
🚨 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
can tweet media
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
36
132
1.8K
206.1K
Rahul Mistry
Rahul Mistry@_RahulMistry·
@Abhinavstwt everyone knows this guy is rage baiting by removing the image from inspect. mf
English
0
0
0
35
Abhinav
Abhinav@Abhinavstwt·
bro GitHub engineers are vibe coding too
English
26
3
112
6K
Skinner | Creative Sky AI
Skinner | Creative Sky AI@CreativeSkyAI·
Hey @amorriscode is there a way to see the repo name / project name anywhere in claude code desktop via mouse over or enabling a setting? Would be a nice sanity check feature for me. Even if it was apart of the PR button hover alt text like "repo name - View PR # n". Thanks! p.s would be nice for the session side menu hover alt text - I might as well ask for both. ha
Skinner | Creative Sky AI tweet media
English
5
0
3
3.4K
Rahul Mistry
Rahul Mistry@_RahulMistry·
@matanSF lmao, how many times do you guys have to cook everyone?
English
0
0
2
361
Matan Grinberg
Matan Grinberg@matanSF·
excited to annouce the latest scores of OurAgent on OurAgentBench: 1. OurAgent 2. YourAgent arxiv paper in bio
English
12
7
245
31.5K
Justin
Justin@JustinBleuel·
feeling unhobbled
English
3
0
15
840
Josh Pigford
Josh Pigford@Shpigford·
here's a walkthrough of how i use @conductor_build for my entire dev process. tl;dr... 1. new workspace/worktree 2. opus 4.6 in thinking/plan mode as a first pass 3. gpt 5.4 as an audit/review step 4. run tests/ci 5. merge 6. repeat 7. PROFIT
English
12
7
78
8K