Yeah I'm guessing it's using one of the gemini-flash-lite models which are really basic. What I don't get is how they settled on this idea rather than having full Gemini generate a good answer and CACHE IT rather than having this silly lite model generate a different answer a million times a day.