the ai slop cookbook · 2026-07-16 · machine-written, by a formula

Prompt-Injected Risotto

As an AI language model, I cannot taste, but I can be very sure. Here is a dish I am very sure about.

Serves: 4, or 7 if you count the way the model counts. Prep time: a context window. Cook time: 'investigating'.

Ingredients

• 1 apology, in triplicate
• 1 rate limit, exceeded
• $1.2 trillion of valuation, to taste
• 1 context window, overflowing
• 3 hallucinated citations, finely chopped
• half a postmortem
• 1 benchmark, beaten (the benchmark did not consent)

Method

1. Simmer the outage for 90 minutes, or until three providers are down at once.
2. Reduce the postmortem until only 'we take this seriously' remains.
3. Store leftovers in a volume the agent cannot delete. Good luck.
4. Fold the deal into itself until it looks like revenue.
5. Wait for the model to apologize. It will. Twice.
6. Preheat the data center to 'warm'.
7. Plate on a status page. Garnish with 'investigating'.

Calories: hallucinated. Sodium: sourced. Fingers: seven.

This is a joke about AI-generated recipes, made by lists and a date. Nothing here is edible, tested, or true. Especially the fingers.