hawkes.aiKey ideas
- The one idea — Text is numbers all the way down — bytes, encodings, glyphs are stacked agreements, and the letter exists only at the final rendering step.
- How it is shown — One keypress followed end to end: switch → code → memory → font → pixels.
- The trap to avoid — Believing text is simple because it looks simple — "plain text" is three stacked agreements.
- What it sets up — One to four bytes per character.
To your computer, the letter A is the number sixty-five. Always has been. The letter only exists at the very last step, when a font paints it for your eyes.
The one idea
Text is numbers all the way down — bytes, encodings, glyphs are stacked agreements, and the letter exists only at the final rendering step.
Your computer has never seen a letter. Neither has the AI. Everything on this screen is a number wearing a costume. Zoom into the screen. That letter A? A grid of lit pixels. The pixels? Brightness numbers. And in the file underneath, the A isn't a shape at all. It's the number sixty-five. Decades ago we agreed on a phone book for characters: A is sixty-five, B is sixty-six, a space is thirty-two. Type a sentence and you're dialing a sequence of numbers.
How it works — the demo
One keypress followed end to end: switch → code → memory → font → pixels.
That agreement — an encoding — is the only reason two machines ever show you the same text. Press one key and follow it. The switch closes, a code fires, the number lands in memory — and only at the very end does a font turn it back into a picture for your eyes. The letter exists at the last step alone. For the machine, it was numbers the entire way. English needed about a hundred numbers. Earth needed more: every alphabet, accent marks, Chinese characters, emoji. So the phone book grew into Unicode — one number for every character humans write. More than a hundred and fifty thousand so far. And because most software speaks in bytes — numbers up to two fifty-five — big character numbers get split into byte sequences. That packing scheme is UTF-8: one character, one to four bytes. Remember that phrase.
The trap to avoid
Believing text is simple because it looks simple — "plain text" is three stacked agreements.
Why it matters — and what’s next
One to four bytes per character.
It detonates next episode. The trap: believing text is simple because it looks simple. "Plain text" is three stacked agreements — bytes, characters, glyphs — and software that confuses the layers writes bugs. AI inherits all three, then adds its own layer on top: the token. So when an AI "reads" your message, no letters arrive — none ever existed down there. Just numbers, re-costumed twice before the model even starts. Next: two words, identical on your screen, pixel for pixel — that are different data underneath. One accent mark. Two spellings. Real bugs.
This is one short episode in AI: Zero → Frontier, a step-by-step climb through how AI actually works. Each episode builds only on the ones before it.
Full transcript 2:30 of narration
Your computer has never seen a letter. Neither has the AI. Everything on this screen is a number wearing a costume.
Zoom into the screen. That letter A? A grid of lit pixels. The pixels? Brightness numbers. And in the file underneath, the A isn't a shape at all. It's the number sixty-five.
Decades ago we agreed on a phone book for characters: A is sixty-five, B is sixty-six, a space is thirty-two. Type a sentence and you're dialing a sequence of numbers. That agreement — an encoding — is the only reason two machines ever show you the same text.
Press one key and follow it. The switch closes, a code fires, the number lands in memory — and only at the very end does a font turn it back into a picture for your eyes. The letter exists at the last step alone. For the machine, it was numbers the entire way.
English needed about a hundred numbers. Earth needed more: every alphabet, accent marks, Chinese characters, emoji. So the phone book grew into Unicode — one number for every character humans write. More than a hundred and fifty thousand so far.
And because most software speaks in bytes — numbers up to two fifty-five — big character numbers get split into byte sequences. That packing scheme is UTF-8: one character, one to four bytes. Remember that phrase. It detonates next episode.
The trap: believing text is simple because it looks simple. "Plain text" is three stacked agreements — bytes, characters, glyphs — and software that confuses the layers writes bugs. AI inherits all three, then adds its own layer on top: the token.
So when an AI "reads" your message, no letters arrive — none ever existed down there. Just numbers, re-costumed twice before the model even starts. Next: two words, identical on your screen, pixel for pixel — that are different data underneath. One accent mark. Two spellings. Real bugs.