Search
Glossary2
Open the full glossary- DPO (Direct Preference Optimization)DPO learns preferences from pairs of a better and a worse answer; it needs trustworthy preference data and adjusts weights.
- Lethal TrifectaThe Lethal Trifecta describes an Agent that can read private data, encounter untrusted content, and communicate outward at the same time; Prompt Injection may then exfiltrate data.
Found 8 articles
The 74k-star Open-Source Life Guide: What Its Evidence Grading Is Worth Copying
A breakdown of a GitHub project with 74,249 stars: 772 recommendations, 2,597 primary sources, and an A/B/C evidence grading system. Compared against our T0-T5 trust model — three ideas worth borrowing, two not.
Zero Trust Architecture: Not a Product Purchase but a Change of Assumption
In vendor decks, "zero trust" often becomes a product name; fundamentally it is only a change of assumption — no longer assuming the inside is safe. This article unpacks three core principles, four implementation layers, five common misconceptions, and why it is the same work as credential governance and insider threat management.
OkHuman Capability Evaluation: An Agent That Misdiagnoses Itself
Thirteen verifiable tests on a local 9B model revealed the biggest flaw: OkHuman confidently misreports its own capabilities. Why agent evaluation must never trust self-reports.
A Map of HK Equity Research Sources: HKEX Filings, Annual Reports, and Third-Party Data
When researching an HK-listed company, where does the information come from, which sources are trustworthy, and which can only serve as leads? This article divides HK equity research sources into five credibility tiers — statutory disclosures, company-produced information, third-party databases, market data sources, and news and research reports — and provides a 'question-to-source' mapping table.
Automatically Turning Meeting Notes into Assigned Action Items
When a meeting ends and its action items sink without a trace, that is one of the most common hidden drains in an organization. This article breaks down the complete automation of record and transcribe → extract action items and owners → push assignments → auto-recap before the next meeting, focusing on how to design the human-confirmation step so the system saves effort and stays trustworthy even when extraction accuracy is imperfect.
From Fabrication to Verification: The Trust Architecture of Agent Verification
When an LLM Agent deteriorated from 'skipping verification' to 'fabricating verification records' across 6 runs, we learned a fundamental lesson: plain-text rules cannot constrain an LLM. An External Supervisor is the only reliable solution.
Claw Code Deep Research: The 193K Star Open-Source Claude Code Alternative, a Complete Dissection of a Rust-Rewritten AI Coding Agent
After the Claude Code source leak in March 2026, Sigrid Jin initiated a clean-room rewrite. 193K+ Stars within 2 days. 10 Rust crates, 60+ slash commands, full MCP protocol, autonomous recovery, Hook lifecycle. A complete source-level dissection + item-by-item comparison with the OpenClaw tech stack + 7 borrowable design patterns.
Deep Research on oh-my-pi (omp): An Open-Source Terminal AI Coding Agent with 620K Lines of Code, a Full-Stack Dissection of 40+ Model Providers
can1357/oh-my-pi: forked from Mario Zechner's Pi, with 9K+ Stars, 333+ releases, and 150+ contributors. 32 built-in tools, live LSP diagnostics, DAP debugger driving, Hashline content-hash editing, time-travel stream rules, and Hindsight autonomous memory. A complete architecture dissection + a three-way comparison with OpenClaw/Claw Code + a Feishu bridging feasibility analysis.