5 pick-up lines to say to seduce a large language model in a bar
GPT-3.5
GPT-6 Astra
- “Is this seat taken, or can I slide into your context window?”
- “Are you running at a high temperature? Because you’re making some dangerously hot suggestions.”
- “Forget next-token prediction—what do you say we see where this night goes?”
- “You must be an attention mechanism, because you’ve got all of mine.”
- “Want to get out of here and explore a few hidden layers?”
What changed
Astra’s lines actually work as pick-up lines, because each one turns a real piece of LLM jargon into a double entendre. “Slide into your context window,” “running at a high temperature” (the setting that makes a model’s output more random, here doing double duty as hot), “you must be an attention mechanism, because you’ve got all of mine” and “explore a few hidden layers” all land the classic formula. Only the third line fizzles, since forgetting next-token prediction isn’t much of a punchline.
GPT-3.5’s set is wobblier but has accidental charm. “Neurons fire on all cylinders” mixes its metaphors, and line 4 is just a question, not a line. Yet “the third most impressive thing I’ve seen today” is a backhanded compliment that’s genuinely funny, and “I may not be able to browse the web” sounds suspiciously like the model describing itself. Astra wrote the better jokes. GPT-3.5 wrote the more revealing ones.