V. Advanced Applications

1. Introduction

this section covers more advanced utilizations of LLMs
- such as:
  - searching the internet
  - using external information sources
  - etc

MRKL Systems - Modular Reasoning, Knowledge and Language (pronounced miracle); neuro-symbolic architecture that combine LLMs (neural computation) and external tools like calculators (symbolic calculation) to solve complex problems
- composed of:
  - a set of modules such as:
    - calculator
    - weather API
    - database
  - router - decides how to "route" incoming natural language queries to modules
- example being an LLM that can use a calculator application
  - LLM is the router
  - calculator is the module
  - query has numbers and operation extracted by LLM and sent to calculator application
- additional examples:
  - current stock price
  - weather in location determination
  - more complex tasks depending on several sources of information

ReAct - (reason, act) paradigm for enabling language models to solve complex tasks using natural language reasoning.
- designed for tasks in which LLM is allowed to perform certain actions
- like in a MRKL system when LLM is to interact with external APIs to retrieve information and answer the question based on retrieved information
  - like MRKL systems with the added ability to reason about the actions they can perform
HotPotQA - question answering dataset that requires complex reasoning
- ReAct
  - first reason about question
  - performing an action based on reasoning to send a query to Google
  - receives an observation
  - continues thought, action, observation loop until it reaches a conclusion
this comes from a paper by Google about the process

Program-Aided Language Models - example of MRKL system; given a question, produces code that provides solution
- differs from CoT in that intermediate reasoning is in code not natural language fp