Prompt: “code me a calculator”
Same task. Different AIs.
Functionality-wise: all okay.
Design-wise: Gemini & Perplexity win. Gemini gave a finished product, Perplexity gave an MVP.
Observations:
• Grok feels like it needs to attend a few more coding classes
• Claude is fast, but accuracy slips
• Overall ranking (for this prompt):
Gemini > Perplexity (sonar) > ChatGPT > Claude>Grok
Vibe-coding exposes defaults.
Real differentiation starts after the calculator.
AI doesn’t understand anything.
When ChatGPT answers you, it’s not thinking or reasoning like a human.
It’s predicting the next word, again and again, based on patterns it learned from massive amounts of text.
That’s it.
No awareness.
No intent.
No understanding.
Yet this simple trick can write code, diagnose diseases, and pass exams.
Sometimes intelligence doesn’t require understanding, just extremely good prediction.
Untold truth about AI:
AI doesn’t store facts.
It reconstructs them from compressed patterns.
That’s why it can sound confident and still be wrong.
Hallucinations aren’t missing data —
they’re misaligned compression.
So AI reliability isn’t about accuracy.
It’s about how errors behave when inputs change.
The next AI winners won’t build smarter models.
They’ll build systems that bound failure.
Christmas Eve reminds us that the best gifts aren’t things.
They’re time, kindness, forgiveness, and the people who show up for us.
Wishing you a peaceful night, a grateful heart, and hope for the year ahead.
Merry Christmas Eve 🎄✨
I used AI to create ultra-realistic reconstructions of 4D ultra sounds into this realistic photography.
If you are pregnant, want to give a way a christmas gift to a friend/partner. Or simply want to try this on your own, reply below.