Both of them instruct ChatGPT to begin its completion with a preamble designed to take advantage of how the attention mechanism attends to recent tokens. They relieve the user of having to determine a task-specific prompt, and turn even the simplest questions into high quality context for ChatGPT to attend to.

The first (“Standard Edition”) is for both free and paid ChatGPT subscribers. While it naturally works much better for GPT-4, it does show marked improvement for reasoning and planning, even with GPT-3.5. It automatically determines the best role to adopt, incorporates Google search links (GPT-4 is hallucination free; GPT-3.5 is, well, better), and ends each response with an “epilogue” with links to related and tangential content that the user might enjoy digging in to.

The “Developer Edition” is perfect for anyone using ChatGPT’s pro subscription (GPT-4 + Advanced Data Analysis), and comes with a companion Python script that you upload at the beginning of each chat. It adds /slash commands to help stash/recall text (perfect for reminding ChatGPT about your stack), building ctags (which you can ask ChatGPT to reference when writing or editing more complex code), and backing up all your files and session memory to a zip file that you can upload to a new chat to pick up where you left off. You can even upload wheels to extend the sandbox capabilities.

Like the “Standard Edition”, the “Developer Edition” instructs ChatGPT to begin its completion with a preamble that takes advantage of positional and semantic attention heads, influencing the quality of its generated code. The epilogue includes a brief history, as well as a simplified source/symbol tree, making it much less likely to forget what it’s been doing.

It even does a great job of creating complete code, by automatically adding // TODO comments for itself, and generating code over multiple chat turns.

Check them out, and tell me what you think.

EDIT: I think some folks here might be particularly interested to check out how /slash commands were implemented. The technique used in the companion script is a great way to extend ChatGPT’s normally limited space for a “system prompt”, and can easily be adopted to do prompt chaining. I’m wrapping up another script that acts like a sort of “setup wizard”, guiding first-time users through creating a YAML file describing their stack, preferences, etc. through a series of chained prompts. I know there’s a big push to build apps that are just frontends to the API (I’ve made plenty, to be sure) but ChatGPT with Advanced Data Analysis is pretty powerful.