Raw LLM Responses

I see coworkers dump excel files into Copilot and ask it to do analyses which would otherwise require careful data manipulation and advanced pivots. The results are usually wrong because GPT isn’t doing math. I breaks my engineering heart that we have created an incredibly complicated simulation of human verbal reasoning, running in complex data centers full with silicon computational devices, and that model of human reasoning is applied to mathematical questions, which the human reasoning model then gets wrong, just like humans would. Instead of just running that math directly on the silicon.