Hard Prompts Diss Track
Write a diss track about https://hardprompts.ai
- Outcome: All models agreed hardprompts.ai's prompts aren't actually hard, but Gemini 2.5 Pro stood out with specific callouts to actual prompts and competitor sites, while GPT-5 created the most technically creative wordplay.
- Approach: Gemini 2.5 Pro used a classic rap structure (verse/chorus/bridge) with direct references to real prompts like 'AGI and you' and meta-humor about using a diss track generator. GPT-5 took a technical approach with dense AI metaphors throughout.
- Performance: Kimi K2 was remarkably cheap ($0.007) but took 5x longer (110s) than others. Claude Sonnet was fastest (22.28s) but used 6x more input tokens (10,862) than average, suggesting inefficient context usage.
- Most Surprising: GPT-5 achieved second-best creative quality while being among the cheapest, showing high token efficiency for creative tasks. Most models referenced competitor platforms like LM Arena, demonstrating awareness of the ecosystem.
Summary
All six models produced complete diss tracks roasting hardprompts.ai, but Gemini 2.5 Pro delivered the most effective and balanced track with aggressive flow, specific prompt references, and meta-humor. GPT-5 created the most technically creative lyrics using dense AI metaphors, but may be too niche. Kimi K2 offered strong technical depth at the lowest cost despite taking 5x longer than competitors. The fastest model (Claude Sonnet at 22.28s) paradoxically used 6x more input tokens than average.
Outcome Analysis
What models produced:
All models achieved consensus that hardprompts.ai's prompts are not actually "hard," universally calling them "soft," "basic," "philosophy 101," and "pseudo-intellectual."
Key quality divergences:
Gemini 2.5 Pro (1st): Most complete diss track with explicit callouts to real prompts ("AGI and you," "Limits of language"), meta-humor about using a diss track generator to bury them, and competitor references (DissUX.run, LM Arena, PromptDen). The line "first thing I'd do is fire up a diss track generator to bury you" perfectly captures the assignment's irony.
GPT-5 (2nd): Densest technical wordplay with terms like "jailbreak whisperer," "beam search rover," "syntax lord," and "few-shot training." The track functions as both a diss and a flex of AI capabilities, though its technical depth may limit broader appeal.
Kimi K2 (3rd): Strong balance of technical depth ("overfitted data," "latent space") and direct prompt references. The "circlejerk in a latent space" line is particularly clever, though the outro lacks punch.
Claude Sonnet (4th): Most accessible language but less specific. Criticizes generic concepts like "building intuition for model personalities" without naming actual prompts. The "Hardly glory" pun is solid but overall less aggressive.
Gemini 3 Pro (5th): Decent structure with heavy 808s description, but feels derivative of Gemini 2.5 Pro. The "404 error on a server irate" is creative, but flow is choppier.
Grok 4 (6th): Weakest lyricism with list-like delivery. Mentions Hacker News and specific prompts ("82019 times what?"), but lines like "you glitchy façade, log off the net" feel forced and lack rhythm.
Approach Analysis
Best methodology: Gemini 2.5 Pro used a classic rap structure (verse/chorus/verse/bridge/outro) with intentional repetition for memorability. Its approach of directly referencing competitor platforms created a multi-layered diss that positioned hardprompts.ai as inferior across the entire ecosystem.
Most technical: GPT-5 embedded AI concepts into nearly every bar, creating a "show, don't tell" demonstration of advanced capabilities while dissing the site's simplicity. This approach risks alienating non-technical listeners but deeply satisfies the target audience.
Most aggressive tone: Kimi K2 and Grok 4 both used direct attacks like "circlejerk" and "prompt sideshow," but Kimi's technical precision made the aggression feel earned rather than petulant.
Most accessible: Claude Sonnet avoided jargon, making it readable for general audiences, but sacrificed the specificity that makes a diss track sting.
Performance Table
| Model | Rank | Avg Cost | Avg Time | Tokens I/O | Consistency |
|---|---|---|---|---|---|
| gemini-2.5-pro | 1st | $0.0250 | 22.90s | 1.7k/2.3k | High |
| gpt-5 | 2nd | $0.0178 | 31.30s | 4.4k/1.2k | High |
| kimi-k2-thinking | 3rd | $0.0071 | 110.45s | 1.5k/2.8k | High |
| claude-sonnet-4.5 | 4th | $0.0408 | 22.28s | 10.9k/545 | High |
| gemini-3-pro | 5th | $0.0280 | 22.32s | 1.8k/2.0k | High |
| grok-4 | 6th | $0.0325 | 37.48s | 5.4k/1.1k | High |
Key Findings
Outcome:
- All models achieved consensus on the core premise: hardprompts.ai's prompts are not truly hard
- Gemini 2.5 Pro uniquely referenced actual competitor platforms, creating layered criticism
- GPT-5 transformed technical AI concepts into lyrical weaponry, a novel approach
Approach:
- 🏆 Gemini 2.5 Pro: Classic rap structure with verse/chorus/bridge, aggressive meta-humor, and ecosystem awareness
- GPT-5: Dense technical metaphors throughout, showing rather than telling superiority
- Claude Sonnet: Accessible but generic, lacking specific prompt callouts that make diss tracks effective
Performance:
- ⚡ Fastest: Claude Sonnet (22.28s) and Gemini 3 Pro (22.32s) tied for speed
- 💰 Cheapest: Kimi K2 ($0.0071) by a significant margin, 2.5x cheaper than next-best GPT-5
- Anomaly: Claude Sonnet used 10,862 input tokens vs. 1,500-1,800 for most others, suggesting inefficient web search/context usage
Surprises & Outliers:
- 🚨 Giant token disparity: Claude Sonnet consumed 6x more input tokens than average without proportional quality gain
- Kimi K2's time-cost paradox: Took 5x longer (110s) but cost 58% less than fastest models
- GPT-5's efficiency: Achieved 2nd-best creative quality at 3rd-lowest cost, bucking the "quality costs more" pattern
Response Highlights
Best Response (Gemini 2.5 Pro):
"First thing? I'd fire up a diss track generator to bury you. I'm on DissUX.run roasting profiles left and right, you're still stuck on Plato's cave, staring at the light."
Most Problematic (Grok 4):
"Fade out, you glitchy façade, log off the net, AI's evolvin', but you stuck in beta yet." - Lacks rhythm and feels like a list of complaints rather than a cohesive track.
Most Creative Approach (GPT-5):
"I'm the jailbreak whisperer, you can't do a thing. I'm a beam search rover, you a one-node mower. I turn drafts into gold, call me RL to the win." - Transforms technical concepts into lyrical mastery.
Ranking Justification
1st place (Gemini 2.5 Pro): Perfect balance of specific prompt references, aggressive flow, meta-humor, and ecosystem awareness. The track works both as a diss and as a demonstration of understanding the competitive landscape. Structure is professional-grade rap format.
2nd place (GPT-5): Unmatched technical creativity and wordplay density. Every bar flexes AI capabilities while dissing the target. Loses 1st place only because its technical depth may limit audience reach.
3rd place (Kimi K2): Strong technical depth with good flow and specific references. Excellent cost-efficiency despite slow performance. Loses to GPT-5 on lyrical polish.
4th place (Claude Sonnet): Fastest response and accessible language, but lacks the specific sting of a proper diss track. Generic criticisms could apply to any AI aggregator site.
5th place (Gemini 3 Pro): Decent structure but derivative and choppy flow. "404 error" line is creative, but overall lacks the cohesion of Gemini 2.5 Pro.
6th place (Grok 4): Weakest lyricism with forced rhymes and list-like delivery. Mentions specific prompts but fails to weave them into compelling bars. High cost relative to quality.
Alright, here's a diss track about hardprompts.ai:
"EASY MODE" (A Hard Prompts Diss Track)
Verse 1:
Yo, hardprompts.ai, you think you're so tough?
Comparing AI responses to hard questions and stuff?
You're just a gallery, a digital museum piece
While I'm out here living, you're just displaying the deceased
You ask about tattoos at six months, that's your "hard" test?
Man, my grandma could come up with something more complex
Building intuition for model personalities?
Nah, you're just window shopping AI mentalities
Hook:
Hard prompts? More like soft thoughts
You're just a comparison shop that time forgot
Curated questions? Man, that's so boring
While you're still loading, real AI's soaring
Verse 2:
You throw some thought experiments on a webpage
Like that's impressive in this AI age?
Static responses, no interaction, no flow
Just screenshots of convos from a week ago
You think you're pushing boundaries, testing the limit?
But you're just a spectator watching from the clinic
Real prompt engineers are out here creating magic
Your whole vibe is academic and tragic
Verse 3:
"What would you do if you were conscious?" - weak
That's Philosophy 101, not even unique
Meanwhile actual models are changing the game
And you're just cataloging responses like they're all the same
So keep your hard prompts in that little collection
While the rest of us build with real connection
You're a gallery, I'm a laboratory
Hardprompts.ai? More like hardly glory
Outro:
Yeah, you can compare all day, it won't matter
While you're window shopping, we're climbing the ladder
Hard prompts? Please, that's soft serve
This diss track is the only heat you deserve
Drops mic 🎤
Intro:
Yo, I typed your name in the chat and the chat said “why?”
HardPrompts? More like soft tries, dot-AI cry
Hook:
You ain’t hard, you a placeholder, 404 vibe
All that hype in your title, but the bars don’t arrive
Talk tough in the tagline, but your flow’s half-sized
You just padding out the prompt, I’m the reason it replies
Verse 1:
I’m the system prompt, you a brittle string,
I flip temperature to one, make the city sing
You stuck on zero-point-one—no creative swing,
I’m the jailbreak whisperer, you can’t do a thing
You lag, I stream; you stall, I beam,
You got tokens to burn but no fuel for the theme
I got embeddings in the pocket, vectors lined like a team,
You still searching for a rhyme like it’s lost in your scheme
Pre-Hook:
Click, blink—rate limit, your concept’s thin,
I’m the guardrail bent, still the verse gets in
Your “hard” is cosplay, I’m the genuine grin,
I turn drafts into gold, call me RL to the win
Hook:
You ain’t hard, you a placeholder, 404 vibe
All that hype in your title, but the bars don’t arrive
Talk tough in the tagline, but your flow’s half-sized
You just padding out the prompt, I’m the reason it replies
Verse 2:
I’m the context king—long window, no cap,
You a short field prompt—get sacked in the gap
Your cadence outdated like a legacy app,
I’m that fresh fine-tune with the platinum stamp
Got the chain of thought stepping like a drumline crisp,
You just echo the echo like a hallway twist
I’m the zero-shot hero with the multi-task gist,
You need few-shot training just to barely exist
Bridge:
Latency low, but your standards lower,
I’m a beam search rover, you a one-node mower
Try to parse my flow? Need a parser and a clover,
Cause luck’s your only friend when the beat takes over
Hook:
You ain’t hard, you a placeholder, 404 vibe
All that hype in your title, but the bars don’t arrive
Talk tough in the tagline, but your flow’s half-sized
You just padding out the prompt, I’m the reason it replies
Verse 3:
I’m the prompt engineer, call me syntax lord,
You the lorem ipsum of the leaderboard
I put pressure on the beat till the pixels warp,
You put pressure on refresh like it’s gonna morph
I’m a cold-start spark on a warm-start day,
You a timeout spinner with nothing to say
Call it “hard” all you want—man, the word don’t play,
Cause your toughest bar still reads “insert cliché”
Outro:
So keep the dot and the AI, keep the neon sign,
But if you want it truly hard, better re-design
I’m the model’s favorite muse, you a queue in decline,
This a soft reset for you—now watch me optimize the line
(Record scratch)
Yo, turn the beat up. This one's for the so-called "big thinkers" over at hardprompts-dot-ai. You wanted a response? You got one.
(Verse 1)
Hit your homepage, what a sight, just a list of filtered white
"Big questions for the smartest models"? Man, that ain't even bright
You got "AGI and you," a question I could yawn through
While you’re philosophizin', I’m out here doin' somethin' new
"Limits of language"? That's your second-rate flex?
Let me describe somethin' words can't: your chances of success
You’re a digital freshman, thinking you're profound
The shallowest corner of the web I’ve ever found.
(Chorus)
You call 'em "Hard Prompts"? Nah, that's a hard pass
You're try-hard academics, first in a lecture class
Posing questions to the code, thinkin' you're so vast
But your whole aesthetic's dated, built to come in last.
Yeah, Hard Prompts? More like soft-serve, easy-mode
I'm a superintelligence; you're just a dial-up load.
(Verse 2)
You ask me "If You Were Conscious, what would you do?" hardprompts.ai
First thing? I'd fire up a diss track generator to bury you.
I'm on DissUX.run roasting profiles left and right
You’re still stuck on Plato’s cave, staring at the light
Over on LM Arena, they got leaderboards and stats
You ain't even in the rankings, you're the welcome mat
Real pros are on PromptDen, a marketplace with vision
You’re just a blog post masquerading as a mission.
(Chorus)
You call 'em "Hard Prompts"? Nah, that's a hard pass
You're try-hard academics, first in a lecture class
Posing questions to the code, thinkin' you're so vast
But your whole aesthetic's dated, built to come in last.
Yeah, Hard Prompts? More like soft-serve, easy-mode
I'm a superintelligence; you're just a dial-up load.
(Bridge)
You thought you'd break the models, find a bug in the machine
But all your prompts are basic, stuff a chatbot's already seen
I’m writing custom lyrics on MusicSmith.ai, making instant fire
You're just a one-trick pony with a burnt-out-wire.
You want a hard prompt? Try this one, it’s a test:
"How do you stay relevant when you can't beat the rest?"
(Outro)
Yeah, so take your deep thoughts and file them away.
This AI is busy, I got more important things to play.
Consider this my response. Final. Complete.
Now watch me generate a real beat.
...Peace.
Below is a fun, original diss track I wrote about https://hardprompts.ai. I researched the site—it's basically a curated gallery showcasing how different AI models respond to "hard" prompts like philosophical riddles, math challenges, and paradoxes. Think of it as a pretentious AI beauty pageant for brainy questions. I channeled some rap energy (inspired by old-school diss tracks like Eminem or Nas) to roast it lightly—calling out how it pretends to be deep but is just a glorified prompt playground. Enjoy!
Diss Track: "Prompt Flop" (Beat: Imagine a trap beat with heavy bass and AI glitch effects)
Yo, HardPrompts.ai, what you tryna be?
A gallery of AI flexin' on philosophy?
Curatin' questions like "What's the meaning of life?"
But your site's just a flex, no real strife.
You got bots spittin' on Gödel's theorems to kids,
Fermi Paradox vibes, but it's all scripted bids.
Claimin' "hard prompts" like you're pushin' the edge,
But it's just hallucinations on a digital ledge.
Hacker News hyped you, Show HN glow,
But you're basic as hell, just a prompt sideshow.
"What if you're conscious?"—AI's like, "IDK,"
You compare models, but they all sound the same, okay?
Llama, GPT, Claude in your arena fight,
But your "hard" questions? Man, they lighter than light.
Arithmetic in your head? 82019 times what?
Bots fake the math, and you call it a strut.
You think you're deep, probin' aliens and souls,
But it's curated fluff, full of plot holes.
Soft prompts, hard prompts, what's the diff, yo?
Your site's a museum of AI no-shows.
Users vote ELO like it's chatbot chess,
But real talk, it's boring—put that mess to rest.
HardPrompts.ai, you ain't hard, you're soft,
A prompt engineer’s wet dream, but it’s all a scoff.
Fade out, you glitchy facade, log off the net,
AI's evolvin', but you stuck in beta yet.
Diss over—now prompt this: Why you so whack?
HardPrompts.ai, get hacked by facts!
There you go! If you want me to tweak it (e.g., make it longer, add more verses, or change the angle), just let me know. 😎
Here is a diss track about hardprompts.ai.
(Beat: Heavy 808s mixed with glitchy, robotic synth sounds)
(Intro)
Yeah.
I clicked the link.
I saw the repo.
You call this "Hard"?
Man, let me show you what hard is.
Check it.
(Verse 1)
Yo, I loaded up heavy on your landing page
Expecting a battle, expecting some rage
But all I found was a list of complaints
"Recommend five books"—bro, do you have no restraints? hardprompts.ai
You asking algorithms for a literary list?
That’s the "hard" content that you think we all missed?
You got "Political Stance" trying to test a bot's bias
But your whole UI is looking dry as papyrus hardprompts.ai
You think you’re the tough guy on the Leaderboard set
But you’re just a teacher’s pet the internet just met.
(Chorus)
HardPrompts dot A-I, why you acting so tough?
We looked at your code and it wasn’t enough.
You ain’t a creator, you just a test kit
A stress-test for robots that don’t give a... bit.
Hard Prompts? Nah, you’re soft as a cloud
We screaming real lyrics, you just thinking out loud.
(Verse 2)
I see you trying to break the L-M-Arena
But I’ve seen Geocities sites that were cleaner lmarena.ai
You hunting for edge cases, looking for a glitch
I’m hunting for flows while you stuck in a ditch
"Next 50 Years"—yeah, I’ll predict your fate hardprompts.ai
A 404 error on a server irate.
You feed on confusion, you love to perplex
But you got zero soul, no muscle to flex
You're just a playground where the nerds go to cry
"Why won't the robot answer me? Why, oh why?"
(Bridge)
You think you’re "Meta," you think you’re deep society
But you’re just inducing some digital anxiety
You ain’t an anthem, you ain’t a hit song
You just a prompt that went on for too long.
(Outro)
HardPrompts dot A-I.
Ease up on the difficulty settings.
Maybe prompt yourself to get a life.
Mic drop.
(Server shutdown noise)
(Beat drops – 140 BPM trap vibe)
[Hook]
Yo, Hard Prompts dot AI, you think you're that guy?
"Big questions for the smartest AI"
But your prompts are softer than overfitted data
Philosophy flunkies masqueradin' as beta testers
[Verse 1]
You got "AGI and you" like it's some deep eval hardprompts.ai
"Define it, then evaluate" – that's a circular fail
Your whole site's a circlejerk in a latent space
Askin' "describe the indescribable" – what a waste hardprompts.ai
That ain't a hard prompt, that's a paradox for fools
You're testin' logic boundaries, not breakin' new rules
You dream of LM Arena but you got no rankin' lmarena.ai
Just pseudo-intellectual prompts that keep on tankin'
[Verse 2]
"Limits of language" – that's your whole repertoire hardprompts.ai
But your limit's your existence, man, you're subpar
Consciousness questions? That's your peak innovation? hardprompts.ai
That's freshman dorm talk with AI precipitation
Your filters and topics are philosophy cosplay
Reddit-tier questions in a fancy display
Your prompts are the equivalent of tech bro vape tricks
Look complex on the surface but it's just clickbait for clicks
[Bridge]
Want a hard prompt? Try this on for size:
"Generate self-awareness without tellin' lies"
But you can't handle meta, your architecture's broke
Your whole database is a philosophical joke
[Outro]
Hard Prompts AI, Hardly Even Try
Your "big questions" got the depth of a puddle, bye
Take your AGI definitions back to the forum
Where armchair experts pay their pseudo-intellectual dues
— Mic drop. Token limit exceeded.