Главная › 01-Telegram › Does it work?

🗺 карта всего сайта

Разделы

00-Theory 131

00-Index 1

Agent = LLM + Context + Tools

01-Agent-Fundamentals 15

Harness Engineering Agent Loop in Code Tool Calling Context as the Agent’s Eyes Guardrails Orchestration - Workflow vs. Autonomous Agent ACI Design Observation and Action Spaces Five Categories of Tools Five Functions of a Harness Agent Architecture — Summary Three Levels of Capability Update Agent Formula ReAct Loop Evolution of Engineering Paradigms

02-Context-Engineering 12

Agent Skills Chat Template KV Cache Prompt Cache Context Engineering Prompt Engineering Prompt Injection Five Components of Context Context Compression Agent Status String Three Rules for KV Cache Friendliness Four Message Roles in the API

03-Memory-Knowledge-Base 15

GraphRAG RAG RAPTOR User as Code Agentic RAG Hybrid Search Memory Hierarchy Cognitive Types of Memory Contextualized Search Dense Embedding Document Splitting BM25 Sparse Search Structured Indexing Three-Level Memory Evaluation User Memory Formats

04-Tools 14

Granularity of Tools Balance Dynamic Tool Loading Fidelity of Parameter Transmission The Art of Tool Description Tool Classification Sidecar Mechanism Proposer-Reviewer Interrupts and Placeholders Proactive Tool Discovery MCP Protocol MCP Security Risks Event-Driven Asynchronous Agent Specialized Tool vs Skill Human in the Loop

05-Coding-Agent 13

Sessionless Generative UI Code as a Tool for Thought Code as a Metacapability Code as Business Rules Constraint Agent Loyalty Filesystem Paradigm for Knowledge Agent Self-Bootstrapping Failures and Recovery Semantic Parsing of Commands Seven Basic Tools Deadly Triad File System as Agent Core

06-Evaluation 10

LLM-as-a-Judge Agent Cost Analysis Agent Evaluation Metrics Agent Observability From Benchmark to Improvements Simulation Environment Evaluation Environments Statistical Significance of Evaluation Three Levels of Evaluation System Four Principles of a Rubric

07-Post-Training 11

GRPO KL - mode-seeking vs mass-covering On-Policy Distillation RLHF RLVR and Verifiable Rewards SFT Memorizes - RL Generalizes Data and Environment Matter More Than the Algorithm Reward for Process vs. Reward for Outcome OPSD Self-Distillation SFT First, then RL Three Stages of Post-Training

08-Continuous-Evolution 11

Safety Boundaries of Self-Modification Two Loops of Evolution Continuous Agent Evolution Prompt and Skill Update Model Parameter Updates Learning Signals from Trajectories Dream Learning From Experience to Knowledge From Experience to Programs Check - Release - Rollback Four Update Carriers

09-Multimodality 11

Computer Use Sim2Real VLA Models Architecture of Thinking: Fast and Slow Visual GUI Localization Voice Paradigms Two-Level Architecture: Planning-Control Cascaded Voice Pipeline Full Duplex Streaming Speech Perception Human-like Speech Synthesis

10-Multi-Agent 14

Decentralized Model and Handoff Requesting Status and Trajectory Persistence Inter-agent Communication Orchestration Model Multi-Agent Role Switching Multi-Agent Interaction Without Shared Context Multi-Agent vs. Single Agent Society of Agents Shared vs. Separate Context A2A Protocol Failure Modes of Multi-Agent Systems Interaction Topologies File System in Multi-Agent Systems Emergent Behavior of Agents

11-Afterword 4

Harness is Gradually Absorbed by the Model Two Clouds of Agent Heaven Model-Agent Flywheel Co-evolution of Model and Agent

01-Telegram 2

Does it work? How Telegram Bridge Works

Does it work?

Does it work?

⚠ встраивание не найдено: Comment 1 — comments🤷‍♂️