Practical tips, hard-won lessons, and insights on prompt engineering, AI language models, and getting better results from ChatGPT
Most people rewrite their entire prompt when they get a bad response. The fix is usually much smaller than that.
I was writing prompts the way I write search queries. That's a different tool, a different model, and it produces different results.
After two years of asking AI to help with data work, I stopped improvising and settled on one template. Here's what it looks like.
It started as personal notes. I kept getting inconsistent results from ChatGPT and couldn't find a resource that explained why.
ChatGPT is not a deterministic system. Understanding why helps you write prompts that get consistent results anyway.