19 How to Solve AI Skipping Tasks or Being Forgetful

19 How to Solve AI Skipping Tasks or Being Forgetful I had more than 10 tasks, but it stopped after doing only 4. One sentence to solve the problem, say it like this: Look, immediate results; it behaves itself instantly. The core trick is simple: give the AI a quantifiable metric for measurement. 'Find all UI prototype pages of c, count the names and quantity, and optimize all remaining unoptimized pages one by one. Output progress x/y for each completion, and output statistical results after everything is finished.' The reason the above sentence works is simply because I set a metric for the number of pages, clearly aiming to optimize all pages one by one, making it unlikely for the model to slack off. Because I wasn't entirely confident in Gemini's instruction following, I detailed the operational steps and the requirements for outputting statistical results. The logic is the same: language is vague, so AI slacking off and doing less might still seem to meet the criteria. Therefore, you must use precise, non-vague things to constrain the AI, such as...