The Statistical Parrot
from Behind the digital curtain Hosted by Sergio Sanchez
Let's run an experiment. I say the start of a saying and you, without thinking, finish it. "A stitch in time..." Saves nine? Perfect. "Don't count your chickens..." Before they hatch? Excellent. "A penny saved..." Is a penny earned? A hundred points. Congratulations: you just did exactly what ChatGPT does. Predict the next word. No more, no less. When you type to a chatbot, there is no sage on the other end. There is an autocomplete on steroids, trained on mountains of text from the internet, guessing word by word what sounds most likely. And it does it so well that sometimes it looks like it is thinking. And here is the unsettling part. That same mechanism that finishes your proverb can write a poem, summarize a contract, invent a court ruling that never existed, and assure you it is real. Today we open the box and see how it works inside, without a single formula, and along the way find out where it got everything it knows, and who got stuck with the bill. A chatbot does not know what it is saying. It knows what usually comes next.
Embed this episode
No transcript yet.