HammerHound

@9hills Which one? There are at least oh-my-pi protects.

关于 oh-my-pi 和 pi，前者其实蛮好用的，对标 claude code/codex 功能非常的齐全，开箱即用。 两者的关系，大概是 Ubuntu 和 ArchLinux 的区别。 如果想体验 Pi Harness，但是真的不想自己折腾各种插件，就 oh-my-pi 足够。

@manateelazycat Generational token wealth.

梁大圣人真的牛逼啊，全球价格屠夫，谁不服就屠谁 才融了100亿美金，说短期目标不以盈利为主 为了支持梁大圣人，我先充1万再说吧

@Michaelzsguo This is a really interesting experiment! And I think Codex's conclusions aren't too off either.

While DeepSeek is pursuing the goal, my Codex agent and I monitor it in the sidecar and guide or correct it as needed. So I thought I would ask Codex to objectively judge DeepSeek’s capability based on multiple rounds of interaction. Keep in mind, Codex does not know it is talking to DeepSeek. It thought it was another Codex agent. Here is Codex’s evaluation of DeepSeek V4 Pro:

@Teknium Did you use the hermes builtin comic skill for this graphic? I find this style really soothing.

We got BitWarden now, to make it easy to manage your keys, rotate them quickly, and coordinate access with your team.

@victor207755822 Thank you for your service to humankind. I really mean it. If nothing else you have at least delayed the onset of techno-feudalism by 4-5 years. Thanks for saving the world.

💎 "My heart is not a stone; it cannot be turned." 💎 Maybe in another timeline 🕰️🌌, DeepSeek doesn’t exist 🐳, there’s no explosion of open-source models 📂🔓, and no API services that simply chase reasonable profit ⚖️💵. But anyway… I’m just endlessly grateful that in *this* timeline, I can pour my youngest, most alive years into a dream of AGI for everyone 🌍🤖✨. That alone is the greatest happiness of my life. I ask for nothing more. 🫶 It still feels surreal 🤯, like Eren Yeager on a summer afternoon, napping under the shade of a tree 🌳☀️😴💤, dreaming a dream that spanned two thousand years ⏳💭. This world is full of utilitarianism dressed up as dreams 🎭, but we have to trust the power of trust ✨🙏. After all… 我心匪石，不可转也 ——《诗经·邶风·柏舟》， 💎 My heart is not a stone; it cannot be turned. 💎❤️ #DeepSeek #AGIForEveryone #OpenSource

@adiautomates @Teknium Switch to a functional model.

first time facing this issue... @Teknium any idea what it could be and how to fix it?

@ryanvogel Sorry to burst your bubble but Kilo Code already has a patent on opencode.

the AI is getting to a point guys

There are dozens of small reasons, pointing in a general direction of the labs needing to have harness designed to their spec, which requires DeepSeek to build their own harness. At what percent to auto-compact, how to send thinking traces per turn, whether to have many tools or few tools etc. Labs can discover emergent nuances of their models which makes certain common elements of the system prompt redundant as the model has already absorbed it. Other times a model works much better with certain external prompt while the new version is being trained to absorb it. There's so much of stuff if you actually pay attention to detail and want to squeeze the most perf and build the best possible user experience. Chatbot model paradigm is like the building block but it's primitive. You can't think of a seriously useful agentic system with either the model or the harness in isolation but rather they always go together and need to work in tandem. It's not possible to achieve this while relying on an externally maintained harness. What can be done though is to use a standard harness as a foundation and keep the modification limited to a separate layer. But ultimately we might see protocols and standards emerge around harnesses to allow interoperability and reduce the need for everyone to reinvent the wheel, similar to how there are many web browsers but a web dev doesn't need to build for each of them separately.

暴论：deepseek不需要自己的harness，只需要把pi整合好就足够优秀了

@thdxr Aha. Now it makes sense how you're able to run your Go subscription without going bankrupt. You're insider trading!

just got inside info that openai is working on a new model

Ideally you'd want to posttrain your model on a stable harness of your choice which works synergistically with your model and vision, rather than depending on the whim of someone else. Unless we have a standards body ensuring all harnesses follow that spec, labs will build their own or else their model will never work at its best on someone else's constantly evolving harness.

@victor207755822 why.... we have too much? codex, cc, antigravity, hermes agent, openclaw, kimi, etc not against it but why.. i hate moving harness so much. stick to one or help improve existing?

🚀 We’re hiring! DeepSeek is forming a new Harness team to build Code Harness from the ground up—may be you can call it DeepSeek Code or something like this hhh🤣🤣🤣 📍 Based in Beijing. Two roles open: 🧠 Harness Product Manager → app.mokahr.com/social-recruit… 👨‍💻 Harness R&D Engineer → app.mokahr.com/social-recruit… Research meets product—let's build it together. Hit the links and apply directly! 🔥 #DeepSeek #CodeHarness #AI #Hiring #ProductManager #Engineering #Beijing #Referral

M3 wen

@crystalsssup I think they might have invited him in to help salvage Anthropic's developer relations.

huge

@moonagedaydrm9 @gaffe_ @yuhasbeentaken Up until R1 paper came around last January. Every frontier lab in the West has almost certainly adopted stuff from it wholesale. There are far more interesting papers coming out of China the last year or so than the US.

@gaffe_ @hammerhoundai @yuhasbeentaken Why cope? All the biggest advancements and most of the most important papers in the current paradigm were done by Americans or in the west.

@atmikaw_chii @yuhasbeentaken Not at all. Have you used the GLM-5.1 model? As far as the usefulness goes it's better in almost every single way. Benchmarks reflect that too. V4 has a great architecture but it's severely undercooked, almost POC. If you like V4 I have no words to tell what you'll think of V4.2.

@hammerhoundai @yuhasbeentaken no? deepseek is better than both

Great question. As it stands, ofc not. But we'll only find out by ~2028. One could argue that in Chinese modernization course AI was the last thing where they started from a disadvantageous position. Like there's no excuse to not be near the forefront of any new major tech going forward. It's no longer 90s or early 21st century where they had only started industrializing. There's a large middle class and mature market now. That's why I think the next 2-3 years are going to be extremely educational about how the rest of this century will play out.

@hammerhoundai @yuhasbeentaken Can they ever win if they’re just copying exactly what’s happening over the pond, just six months later?

One config I'd like (I'm not sure if it's actually a good idea, just started diving into the slock-like paradigm last night, I do feel this is potentially very handy): currently in multi-agent Matrix channel agent is only notified when it's mentioned, what about being notified with the recent messages on every new message or interval (eg every 5 mins) and only being able to send message in the channel using a tool call, instead of directly sending its final response to the channel. Kinda like humans, being able to decide not to say something where it's not needed but staying constantly informed and able to chime in wherever valuable.

@hammerhoundai @yanhua1010 @istdrc Yes last I heard he was working on integrating hermes ;] What did you have in mind for matrix?

突然发现之前折腾什么OpenClaw，Hermes全他妈浪费时间。 这才是真正的AI Native，目前体验下来最舒服的产品

@Teknium @yanhua1010 Hey @Teknium have you looked at the slock project (from @istdrc ) btw? I've been exploring something similar using Matrix channel in hermes agent just to experiment with some ideas. But I think it'd a great fit with the kanban feature.

@yanhua1010 Did you make this product yourself sir

@ryanvogel Isn't this yet another CLI harness? What's better about it compared to say Claude Code?

grok build is pretty good after using it a bit it’s not perfect but it definitely has very big potential after some more iterations

@9hills It's called steering. Kimi CLI and Codex have this too.

pi 有个功能我很喜欢，当Agent在运行时，你再给他发消息，既不会打断运行，也不会排队到Agent运行完毕。 而是在Agent下一次tool call之前插入，这样可以灵活的给一个long-running的agent 注入指令。 比如我这个主Agent老是要自己写代码，我就给他发个规则：禁止主Agent自己写代码和做测试。