How ChatGPT Builds an Answer

Create a 30-second vertical video (9:16) in a doodle color pop style: hand-drawn black ink line art on a clean off-white background, with bright pops of yellow ( #FFD60A) and coral ( #FF5A5F) used only on the key element in each scene. Lines should look like they’re being drawn live, with a playful, bouncy feel. Upbeat lo-fi background music, light pencil-scratch and “pop” sound effects. Scene 1 (0 to 4s): Hook A doodle laptop shows a chat window. A big coral question mark pops out of the screen. On-screen text: “ChatGPT doesn’t ‘know’ the answer.” Voiceover: “ChatGPT doesn’t actually know your answer. Here’s what it’s really doing.” Scene 2 (4 to 10s): The input A doodle hand types: “The best food in Delhi is...” Each word drops into a yellow box. Voiceover: “You type a sentence. It breaks it into small pieces called tokens.” Scene 3 (10 to 18s): The guess A doodle brain made of tangled lines lights up in yellow. Three word bubbles float out with percentage tags: “chole bhature 62%”, “momos 25%”, “pizza 3%”. The highest one glows coral and pops. Voiceover: “Then it guesses the most likely next word, based on billions of sentences it has read.” Scene 4 (18 to 25s): The loop The chosen word slides into the sentence. The brain lights up again, picks another word, and repeats faster and faster until a full paragraph draws itself. Voiceover: “Then it does it again. And again. One word at a time, until your answer is done.” Scene 5 (25 to 30s): Payoff + CTA The paragraph shrinks into a doodle lightbulb that pops yellow. “AI with Shivam” appears in hand-lettered style underneath. On-screen text: “Smart guessing, not magic.” Voiceover: “It’s not magic. It’s really good guessing. Follow for more AI, made simple.