Second Thought
My website is a space for thinking. It contains thinking content alongside the usual articles.
Second Thought is a tool that helps me do it better. The tool is my article harness around the LLM.
What is Second Thought?
Second Thought is a tool that helps me to check my articles and improve their quality.
It currently has two main features:
- Language check: helps to correct grammar.
- Editorial review: provides feedback about the article's content and structure.
Why did I want my own harness?
Firstly, I was learning about AI, LLMs and harnesses. I wanted to create my own harness and try it myself.
Secondly, I wanted a tool that only provides information and tips for me.
Harness of Second Thought vs. a general agent
In the diagram below, you can see the structure of Second Thought's harness compared to a general agent.
You can see that the input is an article that is first segmented, then processed by the harness. After that, the context is passed to the LLM. The process doesn't stop here. The output of the LLM is then processed by a second harness layer (guard, diff). When the LLM provides the output, the harness can check the output, handle errors and provide feedback to the user.
What I learned during the project
- Stop and give time to thinking
- Quality over features
- Work with AI trade-offs
- Architecture before implementation