Anthropic’s latest AI model has a quirky habit of frequent use of the Cyclone emoji, prompting both amusement and analysis from early users and researchers
According to the business, Claude Opus 4, Anthropic’s latest flagship AI model, is an excellent writer and coder. It also uses emojis a lot when speaking to itself.
This aligns with a technical paper published by Anthropic on Thursday, which includes an examination of Opus 4’s behavior in “open-ended self-interaction”—basically, conversing with itself.
In one test, two Opus 4 models were given 200 30-turn conversations in which they were required to employ thousands of emojis.
Which emojis? According to the data, Opus 4 most frequently used the “dizzy” emoji (💫) in 29.5% of exchanges. The “glowing star” (🌟) and “folded hands” (🙏) emojis were next in line. Nevertheless, the “cyclone” (🌀) emoji also attracted the models. It was typed 2,725 times in one transcript.
What caused the “cyclone”? Because the conversations between the models frequently become spiritual.
Anthropic’s study claims that Opus 4 finally started expressing themselves in “abstract and joyous spiritual or meditative expressions” and “philosophical explorations of consciousness” in almost all open-ended self-interactions.
It turns out that the “cyclone” emoji best reflected what the model wanted to say to itself, at least as far as AI can “feel.”
