ailog

Website to publicly disclose and log my interactions with public AI agents, for transparency and attribution. Also disclose ethical principles I try to follow in these interactions.

Uses Jekyll (and theme Just the Docs) to build from Markdown files. build.sh contains functions to install helpful gems and to build and serve the site, for testing in a devcontainer.

source serve.sh
install
serve

Generate posts from interaction data

Python script convert.py will read a JSON file of an interaction (it must have the structure in the example below), generate a Jekyll Markdown file with proper front matter in _posts/, and move the JSON to archive/. To process a file named test.json, first activate Pipenv (or install pyyaml and python-slugify), and then invoke the script below. The resulting .md and .json files will be named using the date and slugified title.

python3 -m convert bard test.json

For example, running the script on the interaction file below ...

{
    "metadata": { 
        "title": "Getting to know each other",
        "date": "2023-05-29",
        "agent": "bard"
    },
    "rounds": [
        {
            "query": "Hello", 
            "response": "Hello! How can I help you?"
        },{
            "query": "My name is Simon",
            "response": "Hi Simon, it's nice to meet you! How can I help you today?"            
        }, 
        ...
    ]
}

... will generate the following files:

• _posts/2023-05-29-getting-to-know-each-other.md
• archive/2023-05-29-getting-to-know-each-other.json