绝对不要放弃独立思考与批判性思维！如果把科学探索完全外包给AI，也许某一天，《三体》中所说的科学界停滞不前的现象将从小说走向现实。幸运的是，现在这种情况还没有发生——科学星球的开拓与维护，一定是我们人类义不容辞的使命，也是我们人类与生俱来的天赋，更是世界送给我们的精彩！

我们相信且有义务迎接这样的未来：AI的出现，将成为科学界的新黄金时代。所以未来的理想范式是，AI与人类共存。

1. “以人为本”完全局限于人类与机器的关系，代表一种人文立场，是目的不是手段
2. “去中心化”完全局限于工程特征与技术手段，是手段不是目的——我们刻意将我们的“去中心化”与其背后的一切政治口号和意识形态划清界限——信念的纯粹，是我们力量的源泉；模糊这条界线，就是稀释我们的使命

1. Risk-sensitive and confidential scenarios that require fast response: Expert Systems
2. Knowledge graphs should be executable expert systems
3. Train expert systems using the Bayesian philosophy of Case Study, i.e., Supervised Learning, Reinforcement Learning, Context Engineering, or some other optimization techniques, etc.
4. The boundaries of Agents and MLLMs will be broken
5. Help the ESG goals
6. The revival of Algebra and Symbolism

1. Foundation Multimodal Neuro-Symbolic AI Models/Agents
2. Incremental Learning using the Bayesian philosophy of Case Study

为叙述方便并避免歧义，本页中出现的所有“以人为本”与“去中心化”被特指如下。以人为本（Human-Centered）是不可动摇的人文立场与技术设计的根本出发点，完全局限于人类与机器的关系，希望机器永远服务于人类。以人为本是立场上杜绝机器失控的主观基础，有利于把“机器为了机器自身的利益而毁灭人类”的事件扼杀在摇篮之中，从根本上实现价值对齐。去中心化（Decentralized）是一种工程领域的技术手段，彻底服务于群体智能工程系统的稳健性、工程系统平衡能力与工程系统级的自愈能力，特指无需联网，能在端侧训练、部署智能模型或者智能体的技术特性。去中心化一来能够让智能系统拥有极致的工程系统的稳健性、工程系统平衡能力与工程系统级的自愈能力，能够避免因中心故障导致系统彻底瘫痪的问题，充分地利用设备的性能；二来能够极致的尊重用户隐私；三来有助于实现技术的平权，降低技术使用的准入门槛，真正推动技术的普及。有了“以人为本”和“去中心化”的假设后，我们看到，我们人类在经历当下的以人工智能为标志的第四次工业革命之后，有希望以最有活力、最有生命力、最积极的样子，自信而稳健地迈入第五次工业革命。这种可能性，极大地鼓舞了我们在人工智能领域的探索。

我们从内心认可当下人工智能的蓬勃发展与繁荣进步，我们坚信人工智能的长期价值，同时我们也尝试解决人工智能领域的技术挑战。因此，我们在技术路线的设计中（特别是技术路线的第一部分和第三部分，我们将批判继承一路走到现今的AI理论与工程宝藏），尽我们的最大努力吸取当下人工智能领域的精华（这体现在我们对未来趋势的判断中——using the Bayesian philosophy of Case Study, i.e., Supervised Learning, Reinforcement Learning, Context Engineering, or some other optimization techniques, etc.）。我们诚挚地欢迎对该尚在雏形阶段的技术路线内容细节的建设性批判、建设性讨论与建设性工作，为我们共同迈向美好的未来世界，奉献自己的力量~

我们始终坚持通用是一种相对性的动态过程——你面对不熟悉的领域就是表现不好，你唯一知道的是你不知道什么，这是事实。所以绝对静态的通用是一种不可能达到的状态，因此持续进化才是一个智能真正需要的能力。

单个模型（请不要局限于专家系统）对应人类世界的专家，多个模型对应人类世界的专家团队，多个模型集成（神经网络在哲学层面上也是神经元的集成，可以尝试的投票机制比如可信度加权，请参考《原则》一书）决策才是泛化的关键。绝对静态的通用并不存在，人类也不过只是解决了几个自己熟悉的领域，不足以标榜自己“通用”。所以，真正的智能应该能够保持谦逊的态度，并借助集成与团队协作的哲学，实现真正的集体持续学习（Incremental Ensemble Learning）。

任何人工智能，我们坚信，只有🍃以人为本🍃的人工智能，才是人类需要的人工智能。

More Early Versions of Foundation Models/Agents will be released, including hands-on training methods and initial datasets, free for anyone!

Nevertheless, cybersecurity is vital. To hinder our products from being abused and to prevent hackers from exploiting vulnerabilities in this platform to gain control over users, the latest versions of this platform are NOT open.

We are a small team whose members major in Mathematics & Engineering, with limited resources (esp. equipment and computational resources). This is why we consecrate ourselves to the advent of a decentralized world. (Remark: We have defined the term "decentralized" above in Chinese.)

Imagine a situation where your creative ideas in your public papers or your open-source projects are ignored in others' relevant public papers or open-source projects. We understand such frustration. We are committed to giving credit to related public papers (including ArXiv) or open-source projects (including GitHub) within our best efforts. If you believe an omission in our references exists, please bring it to our attention by creating an issue in our project. We will highly value those suggestions that provide a direct methodological foundation, introduce a key concept we build upon, or offer a seminal result in the same specific problem domain.

Welcome to our homepage! We will be glad if you share the same mission and dream with us, so we're here seeking collaboration opportunities, to address the following issues in Foundation Models/Agents together:

1. Lacking interpretability, which makes LLMs hallucinate - being opaque, not auditable, and not thoroughly controllable
2. Low inference ability and hence being not trustworthy (as a consequence of hallucination), especially in some risk-sensitive situations
3. Catastrophic forgetting
4. Long response time
5. Difficulties in local deployment and local training (esp. on CPUs of PCs or on mobile phones), where privacy matters
6. Low data efficiency - a low intelligence density, which is the consequence of a lack of interpretability

All of our contributions are indexed in this project, including the Technical Roadmap Draft: https://github.com/Magic-Abracadabra/Office-Agents-and-Their-Incremental-Learning-Framework

• 技术平权：智能可以在普通个人电脑甚至手机上本地训练和部署——智能不再是少数人的特权
• 完全掌控：数据不出设备，推理过程完全自主掌控——用户甚至可以自己亲手复现这样的智能模型/智能体
• 极致稳健的决策系统：系统不会因中心服务器故障而瘫痪——每个个体都可以发挥自身影响力并在分析阶段提供独一无二的视野，群体即可掌握全局、缩小盲点对冲决策风险，在综合决策与压力测试阶段产生最佳战略——我们追求创意择优，不在乎最佳决策源于何处，我们在乎是否扫清了盲区
• 开放与可控：早期模型和框架开源，促进协作与创新；同时核心安全机制闭源，防止滥用和被控制。用户真正拥有对自己智能的完全控制权，最终决策权和控制权在你手中，人类才是最终的裁判，借助智能扬长避短

• 人类成为AI的“指挥家”、“监督者”和“创意总监”
• 人类负责提出问题、设定目标、进行批判性思考和最终决策；AI确保重复性、流程化的脑力劳动的可靠执行——我们不关心你的种族、学科背景、阶层、学历与出身，我们关心你是否与我们共享愿景——在这里，有意义的工作与有意义的人际关系是人间值得的证据——与我们合作只需一个入口与起点，我们会与你充分交互之后，达成最后的共识
• 我们的全球的、官方的、唯一的社交矩阵：

• 防范“科学外包导致科技停滞”——人类永远有权利修改AI的知识——我们致力于确保人类永远不放弃独立思考和科学探索的使命，AI是探索的“望远镜”和“加速器”，而非替代者
• 可解释的推理过程是透彻理解的学习工具——以独立思考、批判性思维为核心的启迪成长方式，确保人类的文明成果经受得住压力测试，用最严格、最全面、最激进、最透彻的质疑捍卫人类文明健康的可持续发展
• 学习，是有趣好玩的创造力体验之旅
• 我们不关心你的种族、学科背景、阶层、学历与出身，我们关心你是否与我们共享愿景——与我们合作只需一个入口与起点，我们会与你充分交互之后，达成最后的共识

We firmly believe that we shall prepare ourselves for a world of Office Automation, Critical Thinking, and Creativity, which will enable us to enter the era of the fifth industrial revolution. Magic, the so-called "consciousness energy", is the ability of consciousness to control the material world. The automation and modularization of the fourth industrial revolution will lay a solid foundation for this god-like ability. People in the fifth industrial revolution will pay more attention to world exploration (satisfying curiosities about the world) and their influence (demonstrating their values), both in the inner and outer worlds. Consciousness energy will become a new era of technology. At that time, an idea is the only requirement to teleport ourselves to a human base station on Mars. Just going through the spell in your mind, you will turn domestic garbage into energy supply materials. Our world at that time will be as exciting and cool as today's games.

Now, it's of great significance to tackle the problems listed above. Let's work together, you and us, to find a better solution of Machine Learning, bringing Magic to this world! Abracadabra!