We could get there if current LLM's managed to prepare some data and offload it to a plugin, then continue on with the result
* LLM extracts the problem and measurements * Sends the data to a math plugin * Continues its reasoning with the result
* LLM extracts the problem and measurements * Sends the data to a math plugin * Continues its reasoning with the result