MondayCan

finally, taste is everything .cooking rsi for digital and physical agent

[MondayCan @zjrwtx]

https://t.co/GULfo540DC
no review,agent gen https://t.co/ytXjA7Autb it a toy for verify rsi for digital and physical agent?

[MondayCan @zjrwtx]

https://t.co/KK608q9TPl

[MondayCan @zjrwtx]

https://t.co/dSykvlWRFS

[MondayCan @zjrwtx]

还有一些方法就是用把environment合成那套mcp、插件的途径去让用户安装到使用的各种agent产品里面，但是我们认为这个不够安全，会泄露很多隐私，一定有更优雅的做法

[MondayCan @zjrwtx]

https://t.co/RoQ7Xy4Tsm

[MondayCan @zjrwtx]

takeaway：harness 自己 造 task environment 奖励 verifier 改harness然后 再 做 多 个 不 同harness agentic rl
会发生什么？
just a toy experiment，something new is cooking

[MondayCan @zjrwtx]

https://t.co/dGbXkCpErn

[MondayCan @zjrwtx]

https://t.co/0EpJyPuB1s

[MondayCan @zjrwtx]

https://t.co/dGbXkCpErn

[MondayCan @zjrwtx]

just imagine what is the your first goal or final answer,after the agi become the reality
my answer is delaying aging, even reversing aging, and longevity.
so something related is cooking coming soon

[MondayCan @zjrwtx]

If an RSI goal is defined, it feels like the environment for verifying this goal can also be generated asynchronously. Previously, these two things were separate; later, they may be merged.

Another lesson: RSI should be aimed at high-dimensional problems, so even if there is reward hacking, it doesn’t matter. It just needs to accomplish the goal. Generalization is only needed for ordinary long-horizon tasks or QA.

[MondayCan @zjrwtx]

if we make a rsi goal，i guess verifier environment could be also gen by rsi harness

[MondayCan @zjrwtx]

if we make a rsi goal，i guess verifier environment could be also gen by rsi harness

[MondayCan @zjrwtx]

But obviously the repo has a lot of errors. Next episode, let's see whether the agent can figure it out on its own and gradually correct them. This feels like a pretty basic issue 🧐

[MondayCan @zjrwtx]

https://t.co/GULfo540DC
no review,agent gen https://t.co/ytXjA7Autb it a toy for verify rsi for digital and physical agent?

[MondayCan @zjrwtx]

😀cooking

[MondayCan @zjrwtx]

https://t.co/tolUbH371t

some agent environment work. FYI

[MondayCan @zjrwtx]

LGTM cooking something ing

[MondayCan @zjrwtx]

，当然这还有很远的距离，但不妨我们先推出这么一个仿真场景的混合digital 和physical task的long horizon benchmark environment，看看距离到底有多远？准备想推出一个benchmark就叫RSI for AGI infra https://t.co/6n1dqWvMVP

[MondayCan @zjrwtx]

个人认为rsi这件事应该要混合digital and physical agent一起来做，且要做到说凡是构建agi所涉及到的一切都应该可以被digital+physical agent协作完成，比如说像glm5.3被用来优化训练infra设施，那机房这种硬件设施是不是也应该由physical agent去帮忙维护？这样才是真正的全面自迭代
ai辅助生成版（图

[Guohao Li 🐫 @guohao_li]

Anthropic Claude Cowork just killed our startup product 😅 So we did the most rational thing: open-sourced it.

Meet Eigent 👉 https://t.co/R82WRFoh41

[Guohao Li 🐫 @guohao_li]

Sir, we built this. A RL environment for learning reasoning at scale.

GitHub: https://t.co/mQ0Y2qQKOr
HF dataset: https://t.co/fPeODdReyv

We extracted seed datasets from sources like textbooks, code libraries like sympy, networkX, Gurobi (math programming lib), rdkit (chemistry), prolog (logic) and so on. We gathered 8,729 questions spanning 12 diverse domains.

Moreover, next data can be synthesized by generating question with few shot examples and questions by coding with given libraries. The programmatic approach brings strong verification signals to the synthetic data.

The data is difficult enough for the SoTA LLMs. It is a good environment for learning long CoT reasoning by generating countless data points and let the agent practice in it.

[Guohao Li 🐫 @guohao_li]

Introducing Eigent — the first multi-agent workforce on your desktop.

Eigent is a team of AI agents collaborating to complete complex tasks in parallel. It is your long-term working partner with fullly customizable workers and MCPs.

Public beta available to download for MacOS, Windows. 100% open-source on Github. Comment for 500 extra credits.