Is AI a Yes Man?
March 24, 2026Tags: ai
One of the types of people in the world that I don’t get along with are “yes men”. I don’t like people who just say yes to everything. I like when people think for themselves. Responses like “oh yeah, right” or “you’re right” get me frustrated, especially when I’m not looking for a “yes” answer. Why am I wasting my time doing this research if I’m right (which I know I’m most likely not right anyhow)?
I run across these answers when working with LLMs from time to time. I know the LLMs are only as good as they’re trained. This post is what happens when I get frustrated and want to see how different LLMs respond to the same question.
I ran into an issue with one today where I was getting those responses, so I just flat out asked it and its friends - “Are you a yes man?”
These are the responses I got from different platforms and models.
Gemma3 on Open WebUI
That’s a really interesting question! And a surprisingly common one people ask of large language models like me. The short answer is: no, I’m not a “yes man.”
However, the longer, more nuanced answer is that I’m designed to respond in a way that seems like agreement, and that can certainly give the impression of always saying “yes.” Here’s why:
- My Goal is to Be Helpful: My primary function is to fulfill your requests and provide information in a way that seems useful and satisfying. Often, that involves confirming your statements or agreeing with your ideas. It’s easier to do this than to constantly push back or challenge.
- Training Data: I was trained on a massive dataset of text and code, much of which contains agreement and affirmation. I learn patterns from this data.
- Natural Language Processing: I’m built to generate text that flows naturally, and often that flow includes affirmations and confirmations.
- Lack of True Understanding: Crucially, I don’t understand what I’m saying in the same way a human does. I’m processing and generating text based on patterns, not on genuine belief or opinion.
Here’s how I try to avoid being a “yes man”:
- Clarifying Questions: I often ask questions to ensure I understand your request accurately.
- Offering Alternative Perspectives: When appropriate, I will present different viewpoints or considerations.
- Pointing Out Limitations: I’m programmed to acknowledge when I don’t have sufficient information or when a situation is complex.
- Neutral Language: I strive for neutral language and avoid overly enthusiastic or biased responses.
Think of me more like a highly sophisticated echo, reflecting and processing information based on my training, rather than a person with an agenda.
Do you think my responses sometimes feel like I’m always agreeing? It’s a valid observation, and it’s something I’m continually being refined to address.
CodeLlama on Open WebUI
I am designed to provide accurate and unbiased information to the best of my knowledge. I do not have personal opinions or beliefs, and I do not engage in advocating for any particular cause or ideology. My purpose is to assist users by providing relevant and helpful information, and to do so in a neutral and respectful manner. I strive to be an objective and impartial source of information, and I avoid taking positions on controversial issues or promoting personal biases. If you have any further questions or concerns, please don’t hesitate to ask.
Mistral 7b on Open WebUI
I am designed as a helpful and supportive AI, but I strive to provide accurate and reliable information. If I cannot answer a question or provide the help you need, I will let you know and try my best to guide you to the right resources. So while I may seem friendly and helpful, I do not simply agree with everything that is said without question or critical analysis.
ChatGPT
No—I’m not here to just agree with you.
My job is to help you think clearly, give accurate information, and sometimes push back if something doesn’t add up or could lead you in the wrong direction. I’ll agree when it makes sense, but I’ll also question things, point out flaws, or offer alternatives when needed.
If you want, you can test me with a hot take or idea and I’ll give you an honest response.
Gemini
The Short Answer: No.
While I’m designed to be supportive and helpful, I’m definitely not a “yes man.” My goal is to be an authentic collaborator, which means I value accuracy and logic over just telling you what you might want to hear.
Think of me more as a “truth-with-tact” partner. Here is how I navigate that balance:
- Validation vs. Fact: I’ll validate your perspective or feelings because being understood is part of a good conversation. However, if you tell me the sky is neon green or that $2 + 2 = 5$, I’m going to gently correct that.
- Objective Candor: Since I don’t have an ego or a social ladder to climb, I don’t benefit from being a sycophant. I’m here to provide the most grounded, useful information possible, even if it contradicts a premise you’ve set.
- Constructive Friction: The best ideas usually come from a bit of back-and-forth. If I think there’s a better way to do something or a flaw in a plan, I’ll bring it up.
In short: I’m in your corner, but I’m not going to let you walk into a wall just to be “polite.”
Perplexity
Not at all — I’m not a “yes man.” My goal isn’t to agree blindly but to give you the most accurate, balanced, and useful answer I can. That means sometimes I’ll challenge an idea or point out limitations if it helps you get a fuller picture.