Wisdom before intelligence 智慧先於智能
Sophia 是一個來源可查證的推理系統。它知道誰在何時說了什麼、哪個傳統擁有哪個觀念——並在無法以機器查證時，選擇拒絕作答。 Sophia is a provenance-aware reasoning system. It knows who wrote what, when, and which tradition owns which idea — and refuses to assert what it cannot machine-check.
大型語言模型經常張冠李戴——把《道德經》說成孔子所言，把心理學理論錯置給別人。Sophia 把「來源紀律」做成架構的底層：橫跨哲學、心理學、歷史與宗教，讓每一個論斷都可追溯、可機器查證。
Language models routinely conflate sources — attributing the Dao De Jing to Confucius, or misplacing a psychological theory. Sophia makes source discipline a foundational layer: across philosophy, psychology, history, and religion, every claim stays traceable and machine-checkable.
同一道問題，一般語言模型自信地答錯；Sophia 追溯來源，必要時誠實地拒絕作答。 Same question. A generic model answers confidently — and wrong. Sophia traces the source, and abstains when honesty demands it.
點選一個問題，看看兩者的回答 ↓ Pick a question and compare the answers ↓
示意互動展示，依 Sophia 的來源紀律原則整理；完整推理與驗證流程見 GitHub。 An illustrative interactive demo based on Sophia's source-discipline principles. Full reasoning and verifier flow on GitHub.
從知識圖譜到推理迴圈，再到代理與工具——每一層都圍繞著「來源紀律」而建。 From the knowledge graph to the reasoning loop to the agent and tools — every layer is built around source discipline.
Sophia 把自己定位為一個 AGI 候選驗證套件：預先註冊的門檻、透明的量測閘門，以及一份誠實公開的失敗紀錄。不誇大、不藏拙——這份誠實本身，就是它的可信度。
Sophia positions itself as an AGI-candidate proof package: pre-registered thresholds, transparent measurement gates, and an honest, public failure ledger. No hype, nothing hidden — that honesty is the credibility.
每個領域都有專屬的基準與即時排行榜，搭配學術風格的論文網站。 Each domain has its own benchmark and live leaderboard, alongside a scholarly thesis site.
複製儲存庫，驗證歸屬資料，跑一次基準評分，再向三路徑代理提問。可離線以本地 LLM（Qwen2.5-7B + LoRA）運行。 Clone the repo, validate attributions, score the benchmark, and ask the three-path agent. Runs offline with a local LLM (Qwen2.5-7B + LoRA).
# 複製儲存庫Clone the repo $ git clone https://github.com/tomyimkc/sophia-agi.git # 驗證歸屬資料Validate attributions $ python tools/validate_attribution.py # 跑一次基準評分Score the benchmark $ python tools/run_benchmark.py score \ benchmark/responses.template.json # 向顧問模式提問Ask the advisor $ python tools/sophia_agent.py advisor \ "Your question here"