Large Language Models (LLMS) can use tools. At the same time, they can also make tools. As we know, they are good at NLP tasks. They can be reinforced by outside tools to enhance their problem solving skills. It is necessary to choose appropriate tools.

In human civilization, mankind evolved to apply the tools to meet the new challenges. Machines too can follow suit. LLMs As Tool Makers (LATMs) enable the use of reusable tools to cover new tasks. There are two components — tool creation and tool application. LATM may assign work to the most eligible LLM at each step.

GPT-4 is resource intensive. It can create tools. A light weight model such as GPT-3.5 may use tools. It reduces computing costs and improves problem solving skills.

Complex arithmetical problems cannot be tackled by light weight machines such as GPT-3.5. However, more efficient models such as GPT-4 can get the answers, though the inference costs are higher. The stronger model becomes a tool maker, and the weaker one a tool user. Once the tool has been forged, it can be deployed to do quick and effective work.

A dispatcher, a light weight LLM, can be used to know whether existing tools are to be used or new tools are to be made. It leads to real time creation and use of tools.

A society can use LLM-generated tools in an evolutionary way.