The Random Logic of Gemini 1.x
Gemini 2 is ready? Great! Because Gemini Advanced (1.x), at least in Europe, seems to operate on an elegantly simple logic:
class GeminiAdvanced:
... # boilerplate stuff ...
def response_to_user(self, question):
rn = random.random() # Generate a random number between 0 and 1
if rn <= 0.5:
return "I am just a language model; I cannot help with that."
elif rn <= 0.7:
return eliza_1966_by_weizenbaum(question) # A nod to the past
else:
return super_sophisticated_transformer_stuff(question)
No wonder it is so fast…
Of course, tools like Claude or GPT have set a high bar for conversational AI, showing what is already possible. Gemini seems to have potential with its “super_sophisticated_transformer_stuff”—but consistency and sophistication are what can make GenAI truly transformative.