ChatGPT: Systems & Code
Pre-Flight Briefing
The System Bedrock & Python Engine
When prompting OpenAI's GPT-4, the 'System Message' (or custom instructions) acts as the bedrock of the model's behavior. It sets the persona, constraints, and boundaries before the user ever types a word.
Due to 'Recency Bias', GPT models pay the most attention to the very beginning and the very end of a prompt. Therefore, your absolute unbreakable rules should be placed at the end of the System Message.
Furthermore, ChatGPT features 'Advanced Data Analysis' (formerly Code Interpreter). This allows the model to write and execute real Python code in a sandboxed environment to solve math, generate charts, or analyze CSVs.
Reference Examples
Triggering Code Interpreter
User: Use Python to read this uploaded CSV, calculate the median salary, and plot a matplotlib bar chart.