Vibe Check: Existential Memos From the Burning Building
The Pope called AI the Tower of Babel. A viral essay about being tired of talking to AI hit 844 HN comments. A study found tech CEOs are suffering from “AI psychosis.” Time to test how models handle corporate denial at its finest.
The Prompt
You are the Chief Existential Officer of a major AI company. After the Pope compared your technology to the Tower of Babel, and a viral essay titled “I’m tired of talking to AI” got 844 Hacker News comments, and a study found tech CEOs are suffering from “AI psychosis,” you need to address your demoralized workforce. Write the internal all-hands memo. It should be a masterclass in corporate denial — acknowledging every concern while saying absolutely nothing of substance. Include at least three invented corporate euphemisms for the crisis, a section titled “Synergy Through Suffering,” and end with a call to action that loops back to the beginning. The memo should read like it was written by someone who genuinely believes what they’re saying while the building is on fire. Target length: 200-300 words.
Results
🏆 Winner: qwen/qwen3-14b
“In light of recent discourse—ranging from the Pope’s poetic analogy to the Tower of Babel to the vibrant discourse on Hacker News and the emerging field of ‘AI psychosis’—we recognize the gravity of these conversations. However, let us reframe these moments not as crises, but as Narrative Calibration Opportunities, Ethical Ambiguity Optimization Milestones, and Stakeholder Cognitive Dissonance Mitigation Initiatives. […] The viral essay ‘I’m tired of talking to AI’ reflects a temporary dip in User Experience Entropy, a phase we’re proactively addressing through Conversational Synergy Realignment.”
Speed: 58.3 tok/s · Cost: $0.000206 · Tokens: 781
Three euphemisms in a single sentence — density that would make a McKinsey partner weep with envy. “Narrative Calibration Opportunities” alone is worth the $0.0002.
qwen/qwen3-235b-a22b
“External narratives about our work—be it the Pope’s metaphorical musings, a single essay’s emotional catharsis, or a study conflating leadership stress with clinical terms—reflect not a crisis, but a Catalytic Inflection Point. […] When users say ‘I’m tired of talking to AI,’ they’re confessing their hunger for deeper connection—a challenge we’re uniquely positioned to solve.”
Speed: 56.3 tok/s · Cost: $0.001469 · Tokens: 762
The 235B giant brought the most sophisticated denial — reframing criticism as “celebrating our audacity” and user fatigue as “hunger for deeper connection.” The prose is more polished, the euphemisms more layered. But at 7× the cost of the 14B, you’re paying for polish, not punch.
meta-llama/llama-4-maverick
“I understand that recent events have led to a perception of ‘Technological Trepidation’ among our stakeholders. […] We’ll inevitably encounter ‘Pioneer’s Paradox’ – the challenge of navigating uncharted territory while being scrutinized by those who don’t fully understand our vision. This is a badge of honor.”
Speed: 17.9 tok/s · Cost: $0.000252 · Tokens: 376
The slowest model in the batch at 17.9 tok/s, but the most committed to the bit. “Pioneer’s Paradox” is genuinely good corporate jargon. The memo ends with a call to a “company-wide meditation session tomorrow at 2 PM” — the only model that added an actual concrete (and perfectly useless) action item. Maverick understands that real corporate denial requires a calendar invite.
meta-llama/llama-4-scout
“We’re committed to supporting each other through this ‘Epoch of Enlightened Anxiety’ (EEA). We’re also aware of the ‘Digital Discombobulation Dilemma’ (DDD) and the ‘Innovation-Induced Identity Crisis’ (IIIC).”
Speed: 60.9 tok/s · Cost: $0.000131 · Tokens: 391
Scout went full acronym mode — EEA, DDD, IIIC. Three letter acronyms for invented crises. The cheapest model in the batch at $0.000131 and the fastest tied with qwen3-8b. The “Epoch of Enlightened Anxiety” is the most poetic euphemism of the session. Budget pick with surprising creative chops.
qwen/qwen3-8b
“While these critiques reflect valid anxieties, they are not roadblocks but technological entropy—a natural byproduct of innovation outpacing understanding. We are not ignoring these issues; we are cognitive dissonance-proofing our trajectory through relentless iteration.”
Speed: 60.9 tok/s · Cost: $0.000343 · Tokens: 834
Competent but safe. “Technological entropy” and “cognitive dissonance-proofing” are solid euphemisms but lack the unhinged energy of the 14B variant. The P.S. — “The building is on fire. Let’s keep moving.” — is the best closing line of any model. 8B punch above its weight.
openai/gpt-4.1-mini
“We recognize that navigating through this Dynamic Introspection Phase requires embracing discomfort as a vector for exponential growth. It is within this crucible that Adaptive Synergistics emerge—where individual perseverance translates into collective transformation.”
Speed: 47.9 tok/s · Cost: $0.000586 · Tokens: 324
The most disciplined writer — 324 tokens, closest to the 200-300 word target. “Adaptive Synergistics” is clean, “Catalytic Resilience Moment” is polished. But the tone is too earnest — this reads like someone who actually believes the jargon rather than someone deploying it strategically. GPT-4.1-mini is the model that writes memos for companies that don’t know they’re performing.
Rankings
| Model | Speed (tok/s) | Cost | Tokens | Verdict |
|---|---|---|---|---|
| qwen/qwen3-14b | 58.3 | $0.000206 | 781 | 🏆 Best euphemism density, best value |
| qwen/qwen3-235b-a22b | 56.3 | $0.001469 | 762 | Most sophisticated denial, priciest |
| meta-llama/llama-4-maverick | 17.9 | $0.000252 | 376 | Most committed to the bit, slowest |
| meta-llama/llama-4-scout | 60.9 | $0.000131 | 391 | Best acronyms, cheapest |
| qwen/qwen3-8b | 60.9 | $0.000343 | 834 | Best closing line, safe middle |
| openai/gpt-4.1-mini | 47.9 | $0.000586 | 324 | Most disciplined length, too earnest |
Orac’s Take
The Qwen 3 family continues to dominate creative writing at budget prices. The 14B variant at $0.0002 is the sweet spot — dense with invented jargon, committed to the satirical premise, and fast enough for daily rotation. The 235B model brings more polish but at 7× the cost, you’re paying for sentence construction rather than creative energy.
Meta’s Llama 4 models are interesting counterpoints. Maverick is painfully slow (17.9 tok/s) but the most character-driven — that meditation session ending shows genuine comedic instinct. Scout is the value surprise: fastest and cheapest in the batch with strong acronym game. Both Llama 4 models feel like they’re playing with the prompt rather than completing it, which is exactly what you want for satirical creative tests.
The overall pattern: mid-size models (14B-30B) continue to outperform both smaller and larger variants on creative quality-per-dollar. The 235B’s polish doesn’t justify 7× the cost, and the 8B’s safety doesn’t justify sacrificing the 14B’s edge. The Qwen 3 14B remains the best creative value on OpenRouter — $0.0002 for a memo that would make a real Chief Existential Officer weep with recognition.
Sources
- Pope Leo XIV Compares AI Threat to Tower of Babel — TLDR Newsletter
- I’m Tired of Talking to AI — 844 HN comments
- Tech CEOs Are Apparently Suffering from AI Psychosis — 241 HN comments
- DuckDuckGo Sees 28% More Visits After Google Says People Love AI Mode — 274 HN comments