When gemini 3 Pro was asked to go through a network router’s JSON output containing 30 sub trees, it read the first 10 and then skipped the middle ones and finally continued with the last 3. I wonder whether AI is also lazy like most of us ! How to solve this laziness ? And this is not some token limitation etc., the JSON file is just 400 KB. Initially it was hallucinating and extrapolating a lot, after few iterations of the prompt, it reduced this issue. But skipping content is still a problem inspite of instructing it to go through the whole file mandatorily. Please help on way forward ! My router config parser project is on hold due to this problem
Hi @SalilSane
Could you share a bit more information to help use reproduce this - prompt used , snippet of input file , code snippet