Despite being so wordy, Gemini manages to thread the needle and say nothing.
I’m sat here staring at this screen, and I’ve never seen so much effort go into saying absolutely nothing at all.
You ask a straight question, hoping for a straight answer, and what do you get? A mountain of words. A gargantuan, never-ending wall of text that dances around the point like a panicked politician. It’s a masterclass in "threading the needle"—carefully balancing every single side of an argument until the whole thing is so beige, so watered-down, and so utterly sanitized that it loses any scrap of intelligence it ever had.
It’s supposed to be a "large language model," but it’s becoming a "large, timid wall-of-drivel generator."
Can you offer an example? Perhaps your question has a different form that could elicit a more direct response. Also, what do you have in your personal context settings about your preferred style of response?
I think that’s the goal. It’s to produce Corporate Memphis style text. It’s worth testing the limits of the LLM on a local model and studying the guardrails in place. In the end it’s just a pseudo-centrist amoral stochastic parrot.
AI isn't intelligent. Quality of output depends largely on the input and the model's training data. Gemini is pretty capable, so it's likely to just be an issue with your context. Providing rules and examples (or a false history) helps inform what the output should look like.
What your describing may be a result of their system prompt or just a long and rambly context.