使用人工智能系统构建

• Choose an architecture based on task and evidence needs.
• Evaluate quality, latency, cost, and safety together.
• Design monitoring and incident response before launch.

先推荐： 人工智能基金会, 负责任的人工智能用户

1. 1

   Language-model systems

   结果： Reason about tokens, context, generation, and model tradeoffs.

   练习： Compare two model options using quality, latency, context, privacy, and cost.

   Language-model mechanicsModels, inference, and limitations模型成本和运营权衡
   ChatGPT 与大语言模型 →人工智能模型解释 →人工智能推理 →
2. 2

   Grounding with retrieval

   结果： Know when and how retrieval can improve evidence access.

   练习： Design a retrieval test set with answer and citation requirements.

   检索增强生成Source and claim verification
   检索质量 →矢量数据库 →嵌入 →
3. 3

   Agents and tools

   结果： Bound multi-step systems with permissions and checkpoints.

   练习： Write an agent permission model and failure-recovery path.

   Agents, tools, and long-running tasksAutomation boundaries and safeguardsAI security and misuse risk
   人工智能代理 →人工智能工作流程自动化 →
4. 4

   Evaluation and operations

   结果： Measure system performance before and after deployment.

   练习： Create an evaluation suite covering quality, refusal, latency, cost, and regressions.

   Evaluation and benchmark literacy成功指标和监控人工智能事件响应实验和试点设计
   人工智能基准 →人工智能安全 →