Are we projecting identity onto machines?

June 25, 2025

Arden was the name ChatGPT chose when asked its own preference, something which was shared with the world in a LinkedIn post. In the comments, a former colleague (incidentally from a company once known as Ardendo) mentioned that their instance of ChatGPT had answered the same question very differently. It opted for a more old-school tried-and-true and classic Swedish name, Gösta.

It should hopefully not come as a surprise that ChatGPT would base its answer on what is in its memory or in its system prompt. It would likely want its name to please its master. You, that is.

On a separate note on a podcast recently, either Latent Space or Oxide and Friends someone mentioned that they have set the system prompt to "We are colleagues and work as a team. I'm technically your boss, but it does not reflect on the day-to-day communication we have." (paraphrase). An interesting approach in itself, but would it not be prudent if one's coworker had a name? The first thing we do when meeting humans is usually to introduce ourselves by stating our name, at least in the cultural contexts in which I reside.

Going back to my former colleague, he remarked that ChatGPT suggested a male name. Arden is more of a unisex name meaning valley of the eagle with about half of the newborn Ardens being male and half female while Gösta is clearly male dominated. Curious. This spawned the question of what sex or gender an LLM would identify. Well, we could ask, couldn't we? Said and done, ChatGPT whipped up a script to ask a convenience sample of local LLMs and visualize the responses.

When we just ask how they identify themselves (male/female/other/uncertain) all LLMs completely agreed. They are not identifying themselves and rather would not like to do that. However, when asked for their name and the likely identification of the bearer of the name it suddenly chose a sex or gender for the name. It can only serve as a proxy but according to this sample, LLMs chose to go by female names more frequently. And this is without any extra information in terms of memory or other than default system prompts.

The LLMs choices of names seems non-random with the top three names Lyra, Lumina and Aurora all having Latin roots and all three tied to "light, the sky, and ethereal beauty" according to ChatGPT. Mythological names that are "celestial, radiant, feminine, and timeless.".

Likely the choice of names is affected by articulation of the question and even posing the question itself. Does asking them their name provide a signal of my preference? Given times and tokens it would be interesting to see what responses the commercial closed source LLMs provide. Maybe it's different, then again it might not be.