LLMs Handle Natural Language Better Than Expected

I’m actually surprised at how well LLMs handle this.

I too expected them to be literal, unable to understand a prompt where I tell it to “make it cooler”

The LLMs handle it just fine. They make assumptions and try to give you something. You tell it “put the button higher” and instead of being an autistic and asking how how many pixels do you want it, they just raise it by about 20% and show you the results.

I was talking dieting and told it I ate a few slices of cheese, and it simply assumed that was 2-3 and gave me a calorie assumption.

They’re much better at natural language than I ever expected. In fact, some humans are worse at this.