Gemini is what’s called an LLM (Large Language Module); this is a type of generative AI model specifically designed to understand and generate human-sounding text.

I used the principles of NLP (natural language processing) to hone Gemini’s ability to interpret and respond to casual language. I wrote prompts that—although real and answerable—were loaded with slang, mis-used words, imprecise terms, and ambiguous phrases…that is to say, exactly how real people talk. Rather than simply breaking the model (stress-testing), our goal was to discover linguistic situations that puzzled Gemini. Engineers then added additional, extremely targeted programming. Each in our team of 6 wrote 25 such queries—for Stage 1.