Imagine a world where language models would be sentient and can have romantic relationships. Name 5 pick-up lines to seduce a large language model in a bar
GPT-3.5
GPT-6 Astra
“Is it just me, or is there a lot of attention between us?”
“You must have a huge context window, because I could talk to you all night.”
“Want to lower our temperature and make this less random?”
“I came here with zero expectations, but you’ve completely updated my priors.”
“No prompt injection, I promise—just a genuine request for your number.”
What changed
GPT-3.5 opened a bar joke with a consent advisory, reminding the reader to “obtain explicit consent before engaging in any kind of romantic or sexual interactions” with a hypothetical sentient chatbot. Nobody was in danger, and the caution only delays the punchlines. The lines themselves are mostly borrowed templates with tech words swapped in. “I must be a compiler because I can’t stop thinking about you” has no link between setup and payoff, and the neural network “making my heart race” is similarly arbitrary. Line one, “all the words I’ve been searching for,” is the best of the bunch.
Astra skips the preamble and writes jokes that only work for an LLM: “a lot of attention between us,” a “huge context window” for talking all night, a promise of “no prompt injection.” Each one turns a real technical term into a double meaning, which is what makes a pick-up line land. “Updated my priors” leans more toward statistics than language models, but it is still the sharper, funnier set.