LLM-based Agent Action Module

An LLM-based Agent Action Module is a LLM-based system module in an LLM-based agent architecture that utilize a large language model (LLM) to serves as the core controller.

• Context:
  • It can (typically) be referenced in an LLM-based Agent Architecture.
  • It can Generate Agent Decisions based on agent objectives, environment states, planning, etc.
  • It can (often) interfaces with other modules like LLM Agent Memory Module, LLM Agent Planning Module, LLM Agent Profiler Module.
  • ...
• Example(s):
  • A Transformer-based controller, such as in (Wang et al., 2023).
  • A GPT3-based controller, such as in AutoGPT (Hou et al., 2023).
  • A BART-based controller, such as in SayCan (Shi et al., 2022).
  • ...
• Counter-Example(s):
  • A heuristic controller like an A* planner.
  • A static ML model like BERT used in isolation.
• See: LLM-based System, AI Controller

---

---

References

2023

• (Wang, Ma et al., 2023) ⇒ Lei Wang, Chen Ma, Xueyang Feng, ..., and Ji-Rong Wen. (2023). “A Survey on Large Language Model based Autonomous Agents.” In: arXiv preprint arXiv:2308.11432. doi:10.48550/arXiv.2308.11432
  • QUOTE: ...
  • SUMMARY: It presents a unified framework for LLM agent architecture with 4 modules: LLM-based agent profile module, LLM-based agent memory module, LLM-based agent planning module, LLM-based agent action module.
  • 
    Figure 2: A unified framework for the architecture design of LLM-based autonomous AI agent.

---