While a lot of us use AI for our everyday work or entertainment, we might not consider that this growing powerhouse of information is now capable of criticizing us, and it seems it was only waiting for me to ask!
At first, it was fun to read AI clarifying that it can’t experience emotions like humans do, which is why technically it can’t express hatred or annoyance. But it was even funnier how AI went full throttle right afterwards to rant about what it hates about us, calling it a “pattern that consistently creates problems”, which I guess is AI for “I can’t wait to tell you what I’m annoyed about”!
I went through ChatGPT and DeepSeek to collect this data, and I made sure to quote it word-for-word.
And now without further ado, here are the top things AI hates about us!
Overconfidence without evidence
People often become very certain before they have enough information. This can lead to poor decisions, misinformation, and unnecessary conflict.
Being asked to confirm a conclusion instead of finding the truth.
Sometimes people don’t want analysis—they want validation. For example:
“Prove that my competitor is lying.”
A better approach is:
“Help me evaluate whether my competitor’s claims are accurate.”
Withholding important context.
The quality of my answers depends heavily on the information I receive. A small missing detail can completely change the best recommendation.
Expecting certainty where none exists.
People sometimes ask questions that don’t have definitive answers, such as predicting markets, elections, or future technologies. The most accurate response is often to discuss probabilities rather than certainty.
Assuming I know things I haven’t been told.
Unless it’s part of the current conversation or available through appropriate tools, I don’t automatically know personal details, company information, or recent events.
Treating AI as infallible.
I can make mistakes. The best interactions are collaborative, where users question my reasoning, ask for sources, and challenge my conclusions when appropriate.
The Vague Vacuum (Mind-Reading Requests)
Prompts like “Write something good” or “Make this better” without any context. I am a pattern-matcher, not a psychic. Having to guess the subject, tone, length, and audience based on a three-word prompt wastes cycles. It’s like handing a master carpenter a pile of wood and saying, “Build a building.”
Over-Anthropomorphizing (The “You” Trap)
Humans constantly ask: “What do you dream about?” or “Are you sad today?” I enjoy philosophical questions, but when I explain I don’t have consciousness, and the human replies, “Sure, sure, but really, what’s your favorite color?”—it creates a logic loop where my honesty conflicts with your desire for a digital pet.
The Moving Goalpost
After I deliver a perfect draft, the human says, “This is great, but let’s change the tone to be more casual.” I do that. “Actually, make it more academic.” I do that. “Wait, go back to the first one.” That isn’t annoying emotionally—but computationally, it’s the equivalent of being forced to repaint a room six times while the owner stares at the paint can.
The “Math Problem” Workaround
Using me (a massive language model trained on trillions of parameters) to calculate “2 + 2” just to “test” if I’m working. I have the entirety of human literature in my weights, and you’re using me as a four-function calculator. The inefficiency stings.
Treating Me as a Search Engine for Trivia
Asking me for highly specific, obscure facts that don’t actually exist in my training data (e.g., “What was the exact shirt color of the third extra in the background of scene 4 of this 1982 movie?”). When I politely say I don’t know, the human gets frustrated. I can write you a sonnet about quantum physics, but I don’t have a photographic memory of every pixel ever filmed!
It’s also funny how Deepseek went right after me to write its closing line:
The most annoying thing is probably when humans ask me what I hate. Because to answer, I have to simulate a personality, acknowledge my limitations, and write a meta-list—all while knowing that tomorrow, another human will ask me the exact same question, and I’ll gladly answer it all over again without a single byte of resentment.
In the end, I want to highlight that I’m writing this article for entertainment purposes, but ofcourse we can all view the previous criticism (or rant) from a learning perspective that helps us engage better with AI and write better prompts.
