For this weekly AI lab, we gave different prompts to open AI. I used Gemini and Deepseek to generate the poem. First of all, I just typed the basic requirement, “write a haiku about happiness,” to both AIs. Then, I added ”a haiku about an abandoned amusement park at dusk, with a twist in the last line.”
It is undeniable that AI can indeed create very creative poems like human poets and writers. However, the advantages created by AI are that it is not easy for people to empathize with some poems written in the context of ancient, modern, and even the “future world” created by AI. Arriagada and Arriagada-Bruneau (2022) mentioned that AI has unique, novel and peculiar creativity. The robust database and algorithm can combine existing poem imitation and autonomous creativity to write better poems. For example, Germini wrote “The storm that cracks the distant coast, The market’s sudden, sharp decline,
A future unforeseen, embossed.” AI uses poems to analyze the current situation of modern society, using the poems written in combination with modern society mentioned above.
Arriagada, L., & Arriagada-Bruneau, G. (2022). AI’s Role in Creative Processes: A Functionalist Approach. Odradek. Studies in Philosophy of Literature, Aesthetics, and New Media Theories, 8 (1), 77-110.
Poem generated by Germini

Poem generated by Deepseek


write a poem that involves chaos theory
Deepseek: A wingtip fractures the air—
somewhere, a monsoon drowns a city.
The pendulum knows this:
its arc, a riddle of gravity and rage,
swings wide enough to unstitch the sky.
We map the static between stars,
trace fault lines in the static of our hands.
Each fingerprint a cipher,
each heartbeat a variable
thrumming in the dark arithmetic of veins.
The mathematician whispers to the storm:
Let me parse your chaos.
Equations bloom like ink in water,
unfolding into hurricanes, into eyelashes,
into the way a match flickers
before the forest burns.
Fractals coil in the curl of a fern,
in the scream of a gull over oil-black waves.
The universe, they say, is a clock
wound by a god who loves dice.
We count the faces,
pray to the hollows between numbers—
for here, in the spin of a roulette wheel,
in the stumble of a drunkard’s walk,
the veil between chance and law
is threadbare, trembling.
All order is a ledger of accidents,
all chaos a psalm of near-misses.
Hold still.
The cosmos balances on the edge of a coin.
Heads: a supernova.
Tails: your breath, tonight,
as you sleep—
a minor turbulence,
a storm in the lungs of the world.
Do you think due to the size limitation of a haiku that it is near impossible for AI to correctly write one? It seems that LLM’s are so busy with trying to find keywords based off of the prompt, that LLM’s do not fully grasp how to write a Haiku effectively.
I don’t know if those haiku matched your expectations before LLMs generated them. Still, as far as I know, Deepseek seems to generate more words because of the use of the Multi Latent Attention (MLA) layers, and it seems to match human reasoning better than Gemini. On the other hand, the LLM you were using from Gemini is Gemini 2.0 Flash, and in most of the benchmark rankings, Deepseek-r1 is way better in generating words and also has larger maximum output tokens as well.