The "Ghost Names" Phenomenon in LLMs
repligate · x · 2026-07-14
Sharing an observation about promptonyms: certain fictional names (like Dr. Elara Voss and its variants) are repeatedly "thought up" by LLMs, acting like ghost characters lurking within the models.
The original text clarifies that the model isn't truly "picking names." Rather, when generating sci-fi/fantasy/speculative fiction, it relies on its training distribution to output statistically high-frequency, highly usable token combinations. This phenomenon has appeared across multiple models including GPT, Claude, Gemini, Grok, Llama, and DeepSeek, and has spread to AI-assisted books on Amazon, Reddit discussions, and writing apps.
More from Fun
- RL comes full circle in games, and this post leans into the joke — y0b1byte · 2026-07-21
- Meme mocks Google’s AI lead after early Gemini 3.6 Flash outputs look rough — max_paperclips · 2026-07-21
- A parody prompt asks for a planetary crystal factory inventory and parity report — Promptmethus · 2026-07-21
- Developer jokes about fixing a kernel module as model safety blocks the request — yacineMTB · 2026-07-21
- Interactive AI game-theory sandbox maps incentives across the AI economy — cpaik · 2026-07-21
- A deleted NanoGPT PR still propagated into later world-record submissions — yacineMTB · 2026-07-21