Haolai(Anthony)Che

Constantly questioning why it is what it is.
Computer Science Ph.D. candidate @CWRU DB Group

[Haolai(Anthony)Che @ZeroSurega]

Excited to share CRUXpider (v0.1.0), an early open-source research asset discovery tool under the CRUX umbrella: https://t.co/ii5L89uMgW.

CRUXpider helps researchers go from just a paper title, topic, or research area to datasets, code, benchmarks, reading paths, and representative papers for AI4Science workflows.

It currently integrates arXiv, OpenAlex, Semantic Scholar, Crossref, DataCite, OpenAIRE, and the GitHub API, and can be launched with a simple one-command local setup.

This is an early release with much room for improvement, but we’re opening it early for feedback and collaboration.

If this direction resonates, we’d love your thoughts, issues, stars, and contributions.

Contact: [email protected]
CC: [email protected], Dr. Yinghui Wu (@yinghuiwu4)
@dbgroup_cwru @cwru

GitHub: https://t.co/VTxYMnNEPr

#OpenSource #AI4Science #ResearchTools #MaterialsScience #MachineLearning

[Kimi.ai @Kimi_Moonshot]

📎We've uploaded it to arXiv, enjoy!
https://t.co/Hyk8ahw8Tc

[Yangyi @yangyi]

斯坦福大学的这篇论文值得了解👇🏻

他们构建了一个AI智能体框架，从零数据起步，没有人工标注，没有精心设计的任务，也没有任何演示，但它竟然超越了所有现有的自博弈方法。

这个框架名为Agent0：通过工具集成推理，从零数据中释放自我进化的智能体。

它所取得的成就令人难以置信。

你之前见过的所有“自我提升”的智能体都有一个致命的缺陷：它们只能生成比它们已经知道的稍微难一点的任务，所以它们会立即达到瓶颈。

Agent0打破了这个天花板。

关键在于：

他们从同一个基础LLM中生成两个智能体，并让他们竞争。

• 课程智能体 - 生成越来越难的任务
• 执行智能体 - 尝试使用推理+工具来解决这些任务

每当执行智能体变得更好时，课程智能体就会被迫提高难度。

每当任务变得更难时，执行智能体就会被迫进化。

这创造了一个闭环、自我强化的课程螺旋，而且这一切都是从头开始的，没有数据，没有人，什么都没有。

只是两个智能体互相推动，达到更高的智能水平。

然后他们添加了作弊码：

一个完整的Python工具解释器在循环中。

执行智能体学习通过代码来推理问题。
课程智能体学习创建需要使用工具的任务。
所以两个智能体都在不断升级。

结果呢？

→ 数学推理能力提高+18%
→ 一般推理能力提高+24%
→ 击败R-Zero、SPIRAL、Absolute Zero，甚至使用外部专有API的框架
→ 所有这些都来自零数据，只是自我进化的循环

他们甚至展示了难度曲线在迭代过程中上升：任务从基本的几何开始，最终达到约束满足、组合学、逻辑谜题和多步骤依赖工具的问题。

这是我们见过的最接近LLM中自主认知增长的东西。

Agent0不仅仅是“更好的RL”。

它是智能体引导自身智能的蓝图。

智能体时代已经解锁。

[Akshay 🚀 @akshay_pachaar]

Google just dropped "Attention is all you need (V2)"

This paper could solve AI's biggest problem:

Catastrophic forgetting.

When AI models learn something new, they tend to forget what they previously learned. Humans don't work this way, and now Google Research has a solution.

Nested Learning.

This is a new machine learning paradigm that treats models as a system of interconnected optimization problems running at different speeds - just like how our brain processes information.

Here's why this matters:

LLMs don't learn from experiences; they remain limited to what they learned during training. They can't learn or improve over time without losing previous knowledge.

Nested Learning changes this by viewing the model's architecture and training algorithm as the same thing - just different "levels" of optimization.

The paper introduces Hope, a proof-of-concept architecture that demonstrates this approach:

↳ Hope outperforms modern recurrent models on language modeling tasks
↳ It handles long-context memory better than state-of-the-art models
↳ It achieves this through "continuum memory systems" that update at different frequencies

This is similar to how our brain manages short-term and long-term memory simultaneously.

We might finally be closing the gap between AI and the human brain's ability to continually learn.

I've shared link to the paper in the next tweet!

[机器之心 JIQIZHIXIN @jiqizhixin]

This paper from Tsinghua University and Shanghai Jiao Tong University received perfect scores (6, 6, 6, 6) at NeurIPS 2025!

It aims to answer a key question: Does reinforcement learning really make large language models better reasoners?

The authors study Reinforcement Learning with Verifiable Rewards (RLVR) and find that while it improves accuracy for small k, it doesn’t create new reasoning patterns—meaning the base model still determines the upper limit of reasoning ability.

Across six RLVR variants, performance gains plateau, suggesting that current RL setups mainly refine reasoning rather than reinvent it.

Interestingly, it’s distillation, not RL, that shows genuine signs of emergent reasoning.

This research points to the next frontier for truly self-improving large language models.

Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model?

Paper: https://t.co/jOvAn6eGIZ
Page: https://t.co/o951L62oR6

Our report: https://t.co/nnMLg0FC5p

📬 #PapersAccepted by Jiqizhixin

[Haolai(Anthony)Che @ZeroSurega]

@Shenxiao123971 🥕

[Haolai(Anthony)Che @ZeroSurega]

@FuSheng_0306 所以单纯的coding已经很cheap了，人类要做的是设计更好的算法，更优秀的理论，更好的架构，更高效的prompt.

[Haolai(Anthony)Che @ZeroSurega]

A paper about how to read a paper.

[Razia Aliani @RaziaAliani]

Read this paper to know how to 'actually' read a paper!
I've highlighted the key points -> now this 10 min read of 3-pass apprach will change your paper reading technique for good!

𝗞𝗲𝘆 𝘁𝗮𝗸𝗲𝗮𝘄𝗮𝘆𝘀:

— First Pass (5-10 minutes)
Quick scan for bird's-eye view

— Second Pass (up to 1 hour)
Read with greater care

—>> Third Pass below ⬇️

💬 𝗛𝗼𝘄 𝗱𝗼 𝘆𝗼𝘂 𝗮𝗽𝗽𝗿𝗼𝗮𝗰𝗵 𝗿𝗲𝗮𝗱𝗶𝗻𝗴 𝗽𝗮𝗽𝗲𝗿𝘀?

1/2

[Haolai(Anthony)Che @ZeroSurega]

@yyds123888 那年我刚上大一，根本没有一百万，就算有认知也跟我没有关系

[Haolai(Anthony)Che @ZeroSurega]

@mirrorzk 好文采。

[Haolai(Anthony)Che @ZeroSurega]

Vibe coding is a wild experience. You enter a state of pure flow and emerge hours later with a masterpiece... that you don't fully understand, almost like a blackbox.

[Haolai(Anthony)Che @ZeroSurega]

@mirrorzk 唐总这个上不上coinbase？

[Haolai(Anthony)Che @ZeroSurega]

@micoolcho COOL! But that face is a little weird for sure

[Haolai(Anthony)Che @ZeroSurega]

So Humanoid Robot is the True Future.

[CLEAN CAR CLUB @TheCleanCarClub]

China gas station

[Haolai(Anthony)Che @ZeroSurega]

@cryptopainter 可惜人只能活一次，你只能活成你自己，所以不需要活在别人的影子里，我虽然不是币圈的，但是“发偏财也是发财”

[Haolai(Anthony)Che @ZeroSurega]

@BinsaeedRashid No he is not otherwise he'd say he is

[Haolai(Anthony)Che @ZeroSurega]

@stats_feed He told you, He is Satoshi Nakamoto.

[Haolai(Anthony)Che @ZeroSurega]

@cneuralnetwork I work in GNN explanation

[Haolai(Anthony)Che @ZeroSurega]

Good resource, but only learn it when you need it. Don't frac your time and energy on random stuff.

[GitHubDaily @GitHub_Daily]

想系统学习数学知识，网上的教材、视频课程、练习题非常多散落在各个地方，要找到质量高的学习资料特别耗费时间。

恰巧，在 GitHub 上看到了 Awesome Math 这份精心整理的数学学习资源清单，已经获得 10000+ Star。

收集了数学各个知识的优质学习资源，涵盖代数、几何、分析、概率统计、数论等 30 多个领域。

GitHub：https://t.co/4HC9Kkd36H

同时提供多种学习方式，比如视频教程、电子教材，以及练习工具等，甚至还有 MIT、哈弗等名校学生的课程笔记。

所有资源都经过精心筛选，目前还在持续更新当中，想自学数学的同学，这份资料值得收藏一下。

[Haolai(Anthony)Che @ZeroSurega]

@haydendevs Linux