Contents › Chapter 2 How it learned
Hide the next character of a story, then compare the AI's answer with the correct one
View in 3DLet's give one of those quizzes to the AI before learning. What comes after "...an honest wood"? The green tile above the blank tile, "c", is the correct answer, as in "woodcutter". The AI's answer is the reel below. The green line under the arrow is the correct slot, but it is so thin it looks like a single line. Grade: badly wrong.
Why?The grade is "how much percent did it give the correct character?" A wide correct slot means a good guess. A thin one means a bad miss. The AI before learning gave the correct answer almost 0%. Badly wrong.
Learn moreMeasuring how wrong an answer is by the percent given to the correct one is called cross-entropy. In AI Through Signals, from the same series, the AI looks at a handwritten picture and guesses the digit. Here, instead of a picture there is a line from a story, and instead of a digit there is the next character.
View this scene in 3D →AI Through Signals (in Korean)
More in the series
A small AI (a character-level language model with 62,652 dials) was really trained on 40 fairy tales and fables retold for this project (about 376,021 characters), and every number here was calculated by that AI. It is a tiny model for learning; real chatbots are far bigger.