Using AI is like a recipe that rewrites itself while you cook

💬 💬

🩺 Summary

You ask AI for a recipe. Halfway through you ask: 'Can I substitute butter for oil?' AI says: 'Absolutely. Also I have reconsidered the entire recipe. Instead of chicken, try tofu. Instead of baking, try grilling. Actually, forget the recipe. Are you in the mood for Thai?' You now have chopped vegetables and no plan.

📝 Details

LLMs treat every follow-up as permission to reconsider everything. A recipe is sequential. LLM attention is not. When you ask about butter, it re-evaluates the chicken, the oven, the side dish, and your life choices. It means well. It just cannot hold a single thread.
Close the chat. Reopen it. Ask for the recipe again. No follow-ups until the dish is on the table. Every question is a new universe.