The act of distinguishing between human intelligence and LLMs is what allows us to figure out how to make it better. There are still some deep limitations, and to ignore them is a mistake. That doesn't take away from how crazy good they are.

User `flanked-evergl` is asking why do you need a whole new format for this instead of just using json. And the answer is because everyone needs to agree on how to specify the format in the json, which `crote` suggests is as simple as doing `{"schema": ..., "data": ...}`.

There's your format. If you want everyone to agree on the `{"schema": ..., "data": ...}` pattern, you can no longer accept just `[]` as a return type. Every response has to include the `schema` and `data` keys.

No, read my comment again. They could do it but it would absolutely destroy the responsiveness of the app.

If you are passing around lists of structured objects, I need you to send me what those objects are supposed to look like even if the list is empty sometimes. If you say "ok so sometimes I'll give you an object with a schema key and a list of data, other times I will just give you an empty list because I don't feel like it", everyone will hate you.

It's the kind of UI that works if you scroll a specific way. A lot of people probably do. But if you don't, it's all very tedious. I consider that to be poor design.

This is why LLMs do their best work at "leaf nodes", building on existing infrastructure but not designing new patterns on their own.

LLMs can't introspect, reason, or build internal models of the world. You can get very far without that, but there are some subtle ways it will bite you, and it's a fundamental limitation. Hallucinations are one: they are the feature, not a bug.