Table 2 Results on ToH (3-disk problems) using a smaller LLM (Llama3-70B)

From: A brain-inspired agentic architecture to improve planning with LLMs

Model

% solved problems

% invalid actions

Llama3-70B Zero-shot

19.2

33.8

Llama3-70B ICL

12.5

41.4

Llama3-70B CoT

29.2

33.3

GPT-4 ICL

46

12

Llama3-70B MAP

50

2

Back to article page