1.Large models are gradually being replaced by AI agent craze.
2.AI agent is an intelligent being, similar to human five senses, brain, and limbs, that can perceive, make decisions, and act. Large models play a key role in this, enhancing the functionality of the planning and decision-making module.
3.It can be understood that the large model is a steam engine, which is a technological innovation (from 0 to 1); AI Agent is a steam train based on a large model, and it is an engineering innovation (permutation and combination).
4.AI Agent = Memory (M) + Planning Decision (P) + Action (A) + Tool Use (T).
Notes:Memory (m):This is the core storage component of AI**, based on LLM. It is divided into short-term memory (STM) and long-term memory (LTM). Short-term memory is responsible for processing information related to the task at hand, while long-term memory stores historical data and experiences to provide the knowledge base for AI**.
STM: Short-term memory, which is used to process instant information and context.
LTM: Long-term memory, which is used to store and retrieve historical knowledge and experience.
Planning (P):AI uses planning modules to set goals, break down tasks, and develop strategies to achieve those goals. This includes ex-ante planning (** and decision-making) and post-event reflection (evaluation and improvement).
Pre-planning: Possible future action paths and outcomes, and development of action plans.
Afterthought: Evaluate the effectiveness of the executed plan, learn from mistakes, and optimize the strategy.
Action (a):Based on the results of memory and planning, the AI selects and executes the most appropriate action. This may include searching for information, reasoning, learning something new, etc.
Tool Use (t):AI** is able to leverage external tools and resources to augment its capabilities, such as APIs, databases, and software, to obtain the most up-to-date information or perform specific tasks. 5.Some experimental projects and LLM applications of AI Agent, such as Stanford Virtual Town, Auto GPT, GPT Engineer, Baby AGI, etc.
6.What you need to learn: LLM: Transformer, LLAMA; Deep Learning Reinforcement Learning.