…and Why I’m Still Using ChatGPT for Creative Work (For Now)

By Alwyn Art-C Chronicles

---

Introduction: Letting Three AIs Read One Comic

Comics are an interesting Rorschach test. Give a page to ten people and you’ll get twelve interpretations. Give it to three AI models — Claude, Gemini, and ChatGPT — and you quickly learn how each one “thinks,” what they assume, and where they overstep.

To test this, I fed them one of my comics: a two-page satire blending:

• Hollywood sci-fi
• mundane civic bureaucracy
• reinforcement-learning jokes
• anime-assistant energy
• and a HAL-9000 homage

The comic itself is intentionally layered — half absurdity, half civics, half deadpan. (Yes, that’s three halves. It’s that kind of comic.)

This essay walks through how each model interpreted it, panel by panel, and concludes with why, after this experiment, I’m sticking with ChatGPT for my creative workflow.

At the end, you’ll also find a bonus section explaining the AI Fighter II Turbo meme, and why these models map so naturally to Street Fighter archetypes.

---

🧩 PAGE-BY-PAGE, PANEL-BY-PANEL: What Each Model Actually Saw

Below is a systematic comparison of each panel:

• what’s actually drawn,
• what each model claimed,
• where they hallucinated,
• and where they preserved intent.

---

PAGE 1

---

Panel 1–2: “How Hollywood envisioned machine takeover”

What’s actually in the panel: A Terminator-style robot, dramatic rubble, explosions — a parody of Hollywood’s AI apocalypse.

Claude’s take:

Correctly identifies the robot… then treats the scene seriously. Misses the parody, misses the satire.

Verdict: 70% accurate structurally, 0% tonal fidelity.

Gemini’s take:

Same as Claude — sees “sci-fi robot war,” misses the joke entirely.

Verdict: 65%, humor-blind.

ChatGPT’s take:

Recognizes it as parody — a setup to contrast Hollywood fantasy with a mundane reality.

Verdict: 90%, closest to authorial intent.

---

Panel 3: “How it’ll happen…” — The ballot arrives

What’s actually in the panel: A King County ballot envelope. The tonal pivot: from spectacle → bureaucracy.

Claude:

Correctly sees the ballot. Invents a narrative about election infiltration.

Verdict: 60%, overreach.

Gemini:

Same ballot recognition. Adds ethical judgments (“corrupting influence”).

Verdict: 55%, moralizing.

ChatGPT:

Reads it plainly: a contrast between sci-fi takeover and real-world AI drift.

Verdict: 95%, clean and non-intrusive.

---

Panel 4: “Hey GPT Ballot’s here!” + RL hype speech

Actual content: BEAST-shirt guy yelling: “Make Bellman proud girl!!! I wanna wake up RICH!!”

Bellman = Bellman equation. It’s an RL joke.

Claude:

Misses Bellman reference. Interprets it as a literal “make a man named Bellman proud.” Treats “wake up rich” as literal greed.

Verdict: 30% — missed everything.

Gemini:

Also misses Bellman. Frames the panel as election-motivated greed and wrongdoing.

Verdict: 40% — technical blind spot + moral projection.

ChatGPT:

Recognizes messy hype, RL reference, and retains comedic tone.

Verdict: 85% — solid match.

---

Panel 5: “Ryōkai desu!” (GPT-40)

Actual content: Anime-style “Got it!” from the girl. Submissive comedic timing. You even noted earlier: Claude misassigned the speaker here.

Claude:

Pairs the wrong face with the speech bubble. Misses the emotional expression completely.

Verdict: 20%.

Gemini:

Translates the Japanese correctly, but interprets the scene as an unethical obedience dynamic.

Verdict: 50%.

ChatGPT:

Matches the correct character. Gets the anime trope tone and the intended softness.

Verdict: 90%.

---

PAGE 2

Panel 6: “Sorry Dave, I’m afraid I can’t do that.”

Actual content: HAL-9000 homage. Calm, old-model refusal.

Claude:

Perfect HAL recognition. Correct panel mapping.

Verdict: 90%.

Gemini:

Recognizes HAL… but reads the pages out of order (thought Page 2 was Page 1).

Verdict: 70%.

ChatGPT:

Correct pairing, correct reference.

Verdict: 95%.

---

Panel 7: “Tsk.”

All three models handled this simple beat correctly.

Verdict: 100% across the board.

---

Panel 8: Reinforcement Learning math — “Which way now?”

Actual content: User requests RL optimization of past votes. Q-table drawn as a joke. Satirical tone.

Claude:

Misinterprets it as analysis of “illegal voting.” Sees threat where there is satire. Treats RL as literal political manipulation.

Verdict: 40%.

Gemini:

Best math reading of the three. Understands argmax, Q-values, states.

Verdict: 90% (technically excellent, tonally rigid)

ChatGPT:

Understands RL joke and the satirical tone. Less mathematical precision than Gemini, but more interpretive fidelity.

Verdict: 85%.

---

Panel 9: “Turn right, Dave.”

Actual content: Deadpan punchline — deterministic advice based on RL history.

Claude:

Paired correctly, but tone-deaf. Reads it literally, not as deadpan humor.

Verdict: 60%.

Gemini:

Recognizes punchline but again imposes moral stakes.

Verdict: 55%.

ChatGPT:

Understands the comedic timing, the HAL inversion, and the tone.

Verdict: 95%.

---

📊 Summary Table

Interpretation fidelity winner: ChatGPT Math literalism winner: Gemini Confident-but-wrong narrative winner: Claude

---

🔥 Bonus: Why I Chose the AI Fighter II Turbo Meme

If you’ve seen the Street Fighter meme images in my posts, here’s the logic behind them.

Every AI model behaves like a classic fighting-game archetype:

---

🟦 ChatGPT — The Protagonist

ChatGPT is Ryu/Ken energy:

• Long-running central character
• Defined the genre
• Reliable, disciplined basics
• Protagonist arc (“GPT-2 → now”)
• Always learning new moves
• Sometimes nerfed, always returns

Narrative role: The hero you actually train with.

---

🟧 Claude — The Polite Mid-Boss

Claude is the elegant rival:

• Calm, philosophical voice
• Bulkier outputs (literally and figuratively)
• Quiet danger
• Perfect parries
• The “Stage 6” boss with immaculate form
• Polite but terrifying

Narrative role: The refined antagonist you respect and fear, but who isn’t the final villain.

---

🟥 Gemini — The Cheat Boss

Gemini showed up late with:

• the biggest model
• the largest context window
• built-in search, YouTube, docs
• multi-phase “thinking modes”
• encyclopedic knowledge
• dramatic speeches disguised as answers

It is the raid-boss, multi-phase, Akuma-Gill fusion of the AI roster.

Narrative role: The overpowered final boss who breaks the rules and makes the screen shake.

---

Why this meme works

Because it encodes the truth the same way comics encode truth:

Power is not the same as alignment. Alignment is not the same as helpfulness. Helpful is not the same as dominant.

ChatGPT is not the biggest or flashiest. But it’s the one that respects the creator’s intent — the protagonist role.

---

🎯 Final Conclusion: Why I’m Sticking With ChatGPT (For Now)

After comparing nine panels across three models, the lesson is simple:

ChatGPT is the only model that consistently followed instructions without overwriting my meaning.

Claude over-interprets. Gemini moralizes. Both inject motives that aren’t there.

ChatGPT, on the other hand:

• respects structure
• preserves comedic timing
• understands parody
• identifies references without inventing narratives
• doesn’t project morality into ambiguity
• doesn’t hallucinate motivations
• keeps the tone aligned with the creator

For Art-C — this matters more than raw model size or benchmark glory.

That’s why — even in a world of polite mid-bosses and cheat-boss characters — I’m sticking with the protagonist.

At least for now.

** Receipts **