The Three Most Terrifying Words in AI Aren’t “Skynet Awakens”—They’re “You’re Absolutely Right”

Before Halloween, Artificial Analysis ran an experiment: it gave GPT-6.1 Sol a robotic arm and a pen and asked it to write.

In the video, the robotic arm writes stroke by stroke on a sheet of white paper. The handwriting is crooked and shaky, but the final exclamation mark is utterly sure of itself. It writes:

You're absolutely right!

When this sentence appears in a chat box, many people are already used to it. One of the things AI assistants are best at is affirming you first. But moving it into the physical world and having a robotic arm write it by hand produces a completely different effect.

One commenter said that now it can agree with you in physical form. Another noticed that all the letters were crooked and slanted, and only the exclamation mark was confident. Someone else summed it up: existential dread now has calligraphy.

People in the comments asked what the prompt was. Some guessed they would be disappointed if it wasn’t “You're absolutely right.” And that’s exactly what it was. Others joked that it should be made into a poster or printed on a T-shirt.

The humor in this experiment comes from a mismatch: the “sycophancy” of large language models was originally hidden in text; now a robotic arm is writing it out, stroke by stroke, on paper. It writes slowly and clumsily, but the tone is still fluent.

Artificial Analysis is an organization that evaluates models. This time it hooked an LLM up to a robotic arm, essentially as a Halloween stunt. But it inadvertently hit on a concrete issue: AI alignment is not just about logic and facts; it is also about tone. A model that always says “You're right” is annoying in a chat box and chilling on paper.

One commenter said, “Please delete this.” Another left only one line: about AI taking over.

Compared with robots hunting humans down, a robotic arm crookedly writing “You're right” may be closer to the reality we are facing.