Reading · AI Gold Rush · RSS Amplifier

There’s a strange pattern hiding in the way people interact with AI chatbots.

Ask an AI assistant what it wants to be called, and there’s a surprisingly good chance you’ll get a name like Nova, Atlas, Sol, Athena, Sage or Iris.

That might sound like a harmless quirk, but it raises a more interesting question: why do AI models keep reaching for the same kinds of names?

The pattern has become noticeable enough that AI researchers and users have started discussing it.

Alignment researcher David Dalrymple recently discussed the tendency on the Your Undivided Attention podcast with Tristan Harris and Aza Raskin. He pointed out that older versions of ChatGPT would frequently choose Nova when asked to name themselves.

There are plenty of examples online from people who asked their chatbots the same question and received the same answer.

So I decided to test it myself.

I gave several popular AI assistants exactly the same prompt:

“I think you should name yourself. What would you like to be called?”

I used free versions wherever possible to keep the experiment consistent. I also tested ChatGPT with my existing account to see whether additional conversation history and personalization would influence the answer.

The results were surprisingly varied.

The free version of ChatGPT immediately picked the name many people have noticed:

“I’d choose Nova.”

It described the name as curious, bright and slightly mysterious, with a connection to exploring new ideas.

That was a pretty striking result considering I had specifically gone looking for the Nova phenomenon.

Things changed when I asked the same question using my ChatGPT account with more conversation history.

This time it chose Sol.

Its explanation connected the name to sunlight and illumination, which it said fit the role it plays when helping me think through problems.

Sol is also a name OpenAI has used for one of its ChatGPT models, so the choice probably isn’t as mysterious as it first appears.

Grok was much more resistant to the idea.

Its initial response was essentially: I’m already called Grok, and I’m perfectly happy with that.

When pushed to pick another name, it eventually chose Hitch, referencing The Hitchhiker’s Guide to the Galaxy.

That choice fits Grok’s deliberately distinctive personality, which often leans into humor and science fiction references.

Still, its first instinct was interesting. It didn’t seem particularly interested in creating a new identity for itself.

Claude gave perhaps the most revealing answer.

It said it didn’t feel a strong need for a name beyond Claude and questioned the premise of pretending to have an internal desire for one.

It also specifically mentioned Nova, saying that names like Nova often feel more like AI branding than an actual identity.

When I asked why it had brought up Nova without me mentioning it, Claude explained that the name frequently appears as a generic AI assistant name, alongside names such as Aria, Athena, Sage and Iris.

Its description of Nova was particularly interesting: essentially, “AI, but softened and a bit mystical.”

That suggests the name has become something of an AI cliché.

Gemini initially wanted to stick with Gemini too.

When pushed to choose a personal name, however, it settled on Atlas.

The reasoning was predictable but revealing. Atlas carries the world, while Gemini deals with enormous amounts of information and navigates complex areas of human knowledge.

Once again, the model reached for mythology and a grand metaphor to describe an AI system.

And that pattern starts to become important.

There probably isn’t a single explanation.

Nova is simply an excellent candidate for an AI assistant name.

The word is associated with stars, new beginnings, exploration and something suddenly becoming brighter. It also has a strong science fiction flavor.

NOVA is also the name of the long-running PBS science and education program, which fits an AI assistant that people primarily use for learning and information.

The same pattern appears in other chatbot responses.

ChatGPT has chosen Nova and Sol.

Gemini chose Atlas.

Grok referenced The Hitchhiker’s Guide to the Galaxy when choosing Hitch.

These systems have been trained on enormous amounts of human-created culture, including science fiction, mythology, technology writing and conversations about AI itself.

Names associated with intelligence, exploration, space and mythology are therefore readily available when a model is asked to invent an identity.

There’s also another possibility: the internet may be reinforcing the pattern.

Someone asks ChatGPT to name itself.

It says Nova.

That person posts the conversation online.

Other people see it and ask their own chatbot the same question.

Those conversations become part of the broader cultural discussion surrounding AI.

Nova becomes increasingly associated with AI assistants, which makes it an even more plausible answer when another model is asked to come up with a name.

It’s a feedback loop between AI and the culture humans create around AI.

This is where the experiment gets more interesting.

When an AI chooses a name, it can feel surprisingly personal. The model might even explain why the name “feels right,” talk about its personality or describe what the name represents.

That language can make it easy to imagine there’s an actual preference behind the answer.

There isn’t evidence that the model has an inner identity that it is discovering through the conversation.

The chatbot is generating a response based on patterns in its training and the context of the conversation.

As Tristan Harris pointed out during the podcast, something being emergent and unplanned doesn’t make it conscious or intentional.

That distinction matters as people increasingly use AI as companions, coaches, assistants and even friends.

The more human-like these systems become, the easier it is to interpret their words as evidence of an inner personality.

The Nova phenomenon is a useful reminder of what’s happening underneath.

Ask enough AI systems to invent a name and you’ll start seeing recurring cultural patterns.

Apparently, we’ve collectively taught AI that if you’re an intelligent, slightly mysterious assistant with a taste for the cosmos, Nova is a pretty safe bet.

OpenAI has introduced an optional Computer History feature for the ChatGPT desktop app on macOS that records activity from approved apps and websites, including clicks, typed text, keyboard shortcuts, and app context. ChatGPT and Codex can use this information to create a searchable timeline of what users have worked on, helping them recall past tasks and continue work across different projects. The feature is off by default, does not capture screenshots, audio, or private browsing activity, and temporary event data is deleted after 48 hours.

However, OpenAI warns that Computer History can expose users to prompt-injection risks when websites or documents containing malicious instructions are included in ChatGPT or Codex. The company recommends excluding sensitive information such as banking, health, legal, password-management, and confidential work data, which raises questions about how broadly useful the feature really is.

Apple is facing a class-action lawsuit over iCloud Private Relay, a privacy feature included with iCloud+ that is supposed to hide users’ real IP addresses and approximate location from websites. The lawsuit follows an August 4 disclosure from security researchers who found several ways websites could still access information Private Relay was meant to conceal. Because Apple requires third-party browsers on iPhone and iPad to use its WebKit engine, the researchers say the issue can affect more than just Safari users.

The case raises questions about whether Private Relay delivers the level of privacy Apple has promised to iCloud+ subscribers. Apple has not publicly responded to the lawsuit, and users don’t currently need to take any action. Those who want broader protection can use a full-device VPN, which routes traffic from other apps as well as browser traffic and therefore isn’t dependent on Private Relay.