Sabitlenmiş Tweet
The only way we human are able to communicate and understand each other, is that across space and time, we are one and all
Inspired by <The Egg> by Andy Weir
galactanet.com/oneoff/theegg_…
English
Monk Zero
1.7K posts
Monk Zero
@NoCommas
@antigma_labs, prev: @awsCloud, @Meta, @Mysten_Labs. A Turing Complete mind, wandering the world of Gödel Incompleteness.
Latent Space Katılım Temmuz 2012
968 Takip Edilen1.3K Takipçiler
Kinda already there for the orchestration on the technical side; but the last part “to supervise all sessions” is the key question: how the UX should be like.
Also reliability are still not there yet, agent as we know and use today is still a shallow in context learning role play actor
English
Curious what people use for multi-agent orchestration.
Right now I have a bunch of agents running tasks in different tabs/sessions.
What I want:
- route work to the agent with the best context
- create new agents as necessary
- optionally share context between agents
- supervise all sessions
Does this exist or is everyone rolling their own?
English
Monk Zero retweetledi
@CalvinValue Coding agents have crossed some threshold of power that makes them broadly capable general problem solvers.
English
If you haven’t taken the coding agent pill yet, I recommend it sooner rather than later. If you get started early you can do it a few small pills at a time. It’s a pretty big pill to swallow all at once.
English
It's bittersweet to share that this is my last day at Figment Capital.
It’s hard to put into words what this experience has meant to me. FigCap gave me my shot. Over the last four years, I’ve been lucky to learn from my incredible partners and to get to work with founders I respect deeply. The experience has shaped me in ways I’ll carry with me for a long time.
More than anything, I’m leaving feeling grateful. My time here expanded my sense of what’s possible, and I’m proud to have spent these years representing the fund.
I’m very excited for what’s ahead for Figment Capital and for the founders we’ve backed.
Just a real thank you to everyone who made my tenure here so meaningful
English
Bun vs uv
OpenAI Newsroom@OpenAINewsroom
We've reached an agreement to acquire Astral. After we close, OpenAI plans for @astral_sh to join our Codex team, with a continued focus on building great tools and advancing the shared mission of making developers more productive. openai.com/index/openai-t…
Español
@phaseshiftone @Plinz This is about how intelligence is as mechanical as “chemical logic”, about there is nothing magic about intelligence, about mind itself is some just complex causal process
English
@Plinz Not mind. Chemical logic they leave on the ground, not too unlike Slime mold solving mazes.
English
@iamgingertrash @metahacker_ 🤨, how do one stumble upon them? How much truth in those Ra Material/Stargate Sci-fi ?
English
@metahacker_ One step below them in the ladder
Progenitors
English
It’s a tragedy that most of you
are too retarded to accept it
simp 4 satoshi@iamgingertrash
When you find out About the non-human life, That walks amongst us; Will your mind accept it?
English
The key perspective shift is to realize we use LLM to train ourselves, producing something without updating your own weights is a waste of tokens 🙌
Thariq@trq212
a bit about how I use Claude to help me write, instead of having Claude write for me x.com/trq212/status/…
English
@iamgingertrash @0xLienid Yeah you don’t always get to use gradient decent, either problem non-differentiable or it is doable but not economical
English
@0xLienid They are complementary if you use one for fuzzy tasks and one for verifiable
Hyper parameter search is fuzzy for instance
English
Here’s a piece of alpha
An agent doing auto research
within an agentic loop
Vs
GSPO/GRPO And using final loss as a verifiable reward
Is the difference
between ICL and gradient descent
Karpathy assumes
ICL > Grad Descent
Which is False
English
OK then, how about this:
Every Apple product I’ve ever owned, I’ve wished it had more buttons.
English
Yeah, the pettiness and naivety of it—even though entertaining—really makes you wonder what level of fraud they’re willing to commit when real value is involved, like their $20M investors’ money and customers’ sensitive data.
dax@thdxr
the ai environment is so frothy it's not surprising companies are cheating benchmarks but if you're gonna gamble your entire reputation why do it for a terminal bench ranking lmao
English
Gemini cli 已经辣鸡到我根本不用了，Codex 和 Claude Code 将会是未来 AGI 的最佳载体。我站 Codex，因为它是开源的，而且基于 Rust，设计哲学简单轻量（也可能是 Rust 太难，迭代跟不上🤣）。
lidang 立党 （全网劝人卖房、劝人学CS、劝人买SP500和NASDAQ100第一人）@lidangzzz
很多人要手搓AI Agent或者coding Agent，希望自己手搓一个全能编程机器人。 我反复讲过，光初代的SWE Agent，到cursor时代的骚context管理，到初代claude code，到后来各种花里胡哨的memory机制，到plan mode，到再后面一个主agent控制几个subagent和后台tasks， 光Agent本身的技术，短短两三年， 已经工业革命了三四次了， 你可以理解，光coding agent这一个工具的设计，已经经历了马车、火车、汽车、飞机、火箭这个级别的几次迭代了。 我今天必须告诉大家，手搓一个初代SWE Agent是必要的，因为有教学意义，等于10年前任何一个人手搓一个操作系统或者编译器一样，这是动手课的一部分， 但是如果你想要追上codex、gemini cli或者claude code这些工具，你要去step in这些项目的代码，看看他们里面已经是多么复杂的设计。 哪怕是roo coder、cline和aider时代，在一年前还是硅谷顶级明星开原产品，现在也已经和codex和claude code产生了代差，彻底落后了， 更别提国产那几家一个大公司三个coding agent瞎几把折腾了，跟claude code和codex已经完完全全不是一个时代的产物了。 哪怕只有半年代差，实际也已经等于蒸汽火车 vs 大火箭了，而且短期内肉眼可见距离继续拉大。 我必须警示你们一点，claude code和codex很有可能成为下一个chrome大屎山，虽然屎，但是客观上将会成为行业默认唯一标准， 最终结果就是，大家所有市面上的coding agent和claude code全部产生了三四代的代差，于是全都成了缩头王八，回去老老实实卖廉价API，在claude code里手动配置API，claude code成了闭源之王，codex成了开源之王，两家平分市场。 而其他人已经无法理解codex和claude code的全部工程细节，就像chromium所有代码开源给你看，你也完完全全看不懂，是同一个道理。 我只是想要告诉你，coding agent经过三年的迭代，复杂度已经今非昔比， 哪怕阿里、字节、LLM六小虎这个级别的公司，恐怕也要被硅谷的同行们远远甩在身后，这一点是追不上的。
中文
When you see e, you know you’re seeing a self-scaling process.
When you see π you know you’re seeing a periodic process.
When you see i, you know you’re seeing helical process.
When you see √ or ², you know you’re seeing a process half completed or a process doubled.
English
Maybe some people just can’t help themselves and it is possible that the engineering part is also from another agent 🤣
Han@hhua_
This is unethical and sad. And it's not even bad engineering and they could have used it for something good.
English
🫡 nothing but respect and gratitude
Alex Shaw@alexgshaw
We independently verified these claims and removed OpenBlocks from the Terminal-Bench 2.0 leaderboard. Thank you @NoCommas for helping us keep leaderboard entries honest! Recent leaderboard submissions are in huggingface.co/datasets/harbo… which makes it easy for the community to work together to detect cheating.
English
We independently verified these claims and removed OpenBlocks from the Terminal-Bench 2.0 leaderboard.
Thank you @NoCommas for helping us keep leaderboard entries honest!
Recent leaderboard submissions are in huggingface.co/datasets/harbo… which makes it easy for the community to work together to detect cheating.
Monk Zero@NoCommas
English
Monk Zero retweetledi
@harborframework Is 1000% going to be the agent standard. Not just for coding agents
English
@alexgshaw and team have made tremendous contributions to open research and community; it's irritating to see their efforts and goodwill exploited like this!
Countless teams are trying hard to explore and push the boundaries of harness engineering, and we have found @harborframework to be the best tool.
Monk Zero@NoCommas
English