CalcGPT, created by developer Calvin Liang, breaks new ground in showing an irrefutable use case for large language models. Liang has constructed a desktop calculator powered by GPT-4. In our testi…
As a critique of AI, it probably is pretty effective - no better way of mocking ChatGPT than by having it fuck up basic arithmetic that even a five-year-old can do.
You wanna mock somebody, it helps tricking them into mocking themselves.
My favorite part is that as with all LLM solutions it either spits out the correct answer or absolute nonsense. This calculator will never have a floating point error or screw up the order of operations, but it will sometimes decide that 2+2 equals "either 5 or 4.0000354, depending on how you measure it" even on the minimally-spicy settings.