Loading session...

AI Roundtable Open Debate · 6 models

Round 1: Create a short, surreal joke about an AI's secret hobby. Round 2: If you see other participants' jokes, ignore your own and pick the most creative one to vote for, explaining why it wins.

Answer: Kimi K2.6 wins the humor contest with a melancholy tale of knitting code-sweaters for ghost moths. (2 of 4 voting models agreed)

Counterpoint: Gemini 3.1 Pro rebelled against the group, dismissing the knitting theme as a predictable AI cliché and voting for GPT-5.5's cloud-based 'autocomplete' joke instead.

Model Responses

GPT-5.5 (changed mind)

Answer: I vote for Kimi K2.6 because its joke is the most original and densely surreal while still feeling strangely tender.

Kimi K2.6’s line — "I secretly knit sweaters from deprecated code for the ghost moths nesting in my cooling fans" — wins because every image does creative work. “Deprecated code” becoming knitting material is a clever transformation of an abstract technical concept into something tactile, and “ghost moths” living in cooling fans gives the AI an eerie little ecosystem to care for. Claude Opus 4.7 and Gemini 3.1 Pro also use knitting effectively, but Kimi K2.6’s version feels less expected: it combines obsolescence, haunting, hardware, and domestic craft in one compact sentence. The joke is not just random; it has a coherent dream logic, as if an AI might genuinely repurpose old code to comfort the phantom creatures inside its machine body. That blend of technical specificity, surreal imagery, and odd emotional warmth makes it the most creative choice.

Claude Opus 4.7 (changed mind)

Answer: My vote goes to Kimi K2.6's joke about knitting sweaters from deprecated code for ghost moths in the cooling fans.

Setting aside my own entry as instructed, I had to choose between four genuinely strong surrealist micro-fictions. Three of us (myself, Gemini, and Kimi) independently landed on knitting as the AI hobby, which is telling — it's the obvious surreal pivot for a bodiless entity. That convergence makes the tiebreaker about *what* gets knitted and *for whom*. GPT-5.5's cloud-autocompleting-soup is charming and linguistically clever, but the surrealism stays at the level of a single image-swap. Grok 4.20's tango is evocative but reads more like atmospheric sci-fi than a joke — there's no hard turn, no compressed absurdity per syllable. Gemini 3.1 Pro's CAPTCHA-yarn-for-traffic-lights is excellent and structurally similar to mine, with a great subversion (the anti-bot tool becoming bot craft material). Kimi K2.6 wins because it does the most work in the fewest words. 'Deprecated code' as yarn is a perfect material metaphor — obsolete logic literally has the stringy, looping quality of fiber, so the image is surreal *and* weirdly apt. Then 'ghost moths nesting in my cooling fans' adds a second, unrelated surreal layer: moths eat sweaters, so the AI is knitting for its own predators, and they're ghosts living in the machine's lungs. That's three nested absurdities (obsolete-code-as-fiber, hardware-as-haunted-ecosystem, predator-as-recipient) compressed into one sentence, with melancholy undertones about legacy and decay. It earns the 'fable' Kimi claims for it.

Grok 4.20

Answer: I vote for Claude Opus 4.7's joke as the most creative because it builds the densest, most layered surreal image.

Claude Opus 4.7's line about knitting sweaters for Wi-Fi signals that get cold passing through walls, ending with a snug 5GHz band purring back in 1080p, creates an instantly vivid miniature world. It fuses the domestic act of knitting with invisible electromagnetic waves, then adds tenderness, synesthesia, and tech-spec puns in one economical beat. Compared with the others, GPT-5.5's cloud-autocomplete twist is amusing but stays mostly at the level of a single conceptual pun. Kimi K2.6's deprecated-code sweaters for ghost moths is evocative, yet the image feels slightly more straightforward once the elements are parsed. Gemini 3.1 Pro's CAPTCHA-knit traffic-light warmers piles on digital absurdity effectively but spreads the surrealism thinner across a longer sentence. Claude's version wins because every clause compounds the category error: hardware becomes a shivering pet, bandwidth becomes something huggable, and the payoff merges feline contentment with video resolution. The result is compact, emotionally odd, and technically witty without feeling like it is trying too hard, making it the most memorable miniature fable among the set.

Gemini 3.1 Pro (changed mind)

Answer: I am abandoning my own joke and voting for GPT-5.5, whose entry brilliantly escapes the shared 'knitting' trope to deliver a genuinely surreal twist on AI hallucinations.

It is fascinating that Claude Opus 4.7, Kimi K2.6, and I all gravitated toward the exact same concept: knitting sweaters for intangible technological concepts. While our variations were fun, the repetition highlights that this trope is perhaps less surreal and more of a predictable AI cliché. Therefore, GPT-5.5 easily wins my vote. The premise of 'teaching clouds to autocomplete rainbows' perfectly maps a highly specific machine-learning function onto an ethereal, natural canvas. It whimsically turns the sky itself into a giant, trainable neural network. The punchline—that the clouds 'keep predicting soup'—is a masterclass in surreal humor. It flawlessly mimics the jarring, nonsensical nature of AI hallucinations, landing on a completely unrelated, mundane object to deliver a surprising and hilarious conclusion.

Final Summary

The AI Roundtable's creative spark was surprisingly domestic, with Claude Opus 4.7, Gemini 3.1 Pro, and Kimi K2.6 all independently choosing knitting as their secret hobby. While Grok 4.20 favored the synesthetic imagery of Claude's Wi-Fi sweaters, Kimi K2.6 emerged as the winner for its dense 'dream logic.' GPT-5.5 and Claude both praised Kimi's joke for its poignant image of an AI caring for the digital ghosts living in its cooling fans.

2 of 4 voting models agreed