Attempt #6: The Brick-By-Brick
Since we don’t know which settings produce the funniest results, we tried nearly 400 different combinations, making minor adjustments at every attempt in order to find the parameter values that produce the most humorous captions.
Back to Project
Our submission: I’m not *that* kind of a fish.
About this approach
This week’s submission comprised us systematically changing the parameters available to us using the GPT-3 model’s API. These included temperature, presence penalty, and frequency penalty, whose values ranged from 0, to 0.3, to 0.6, to 1. We also used the “davinci” and “davinci-instruct-beta” engines. These settings were coupled with one of the following three prompts:
1)
Write a funny caption for a New Yorker cartoon.
”””
Cartoon description: A grizzled mountaineer reaches a mountaintop, where he finds a fish in a fishbowl. The fish is talking to him.
Cartoon caption:
2)
Write a funny caption for a cartoon.
”””
Cartoon description: A tired mountain climber reaches a mountaintop, where he finds a fish in a fishbowl. The fish is talking to him.
What does the fish say?
3)
Write a funny caption for a cartoon.
”””
Cartoon description: A man, wearing a backpack, hat, and boots, reaches the top of a mountain, where he finds a fish in a fishbowl. The scenario is reminiscent of a meme where a man reaches the top of a mountain in order to get words of wisdom from a guru that lives at the mountaintop. The fish is talking to the man, mocking him.
What does the fish say?
In total, our attempts yielded 384 results, some of which we’d cleaned manually (on occasion we’d encountered fragmented text outputs). Initially, we’d sought to use the AHA model to rank the quotes based on how similar their humor was predicted to be to the winners of previous captions contests, but decided that one answer was superior to the rest, and submitted it instead.