Lesson 15

Thinking out loud

It gets the answer right by writing out the working first.

Lesson 07 showed you what it gets wrong. Here is the fix that costs nothing but time. A hard question answered in one guess is one enormous gamble. Broken into six lines of working, it becomes six easy guesses, because each line it writes becomes part of what it reads for the next one. Newer models do this automatically and call it reasoning.

One long jump is a gamble. Six short ones are not, because every place it lands becomes part of what it reads for the next.

1Pick a question it finds hard

2Pick how it should answer

Do thisLeave it on Answer straight away, press the button in step 3 and read the result. Then switch step 2 to Think it through first and step through the working. Same model, same question. The only difference is whether it wrote anything down first.

3Watch it answer

chunks written: 0 time you wait: 0.0s cost to run: $0.00000

The timings and costs are illustrative, but the ratio is real: reasoning models routinely write ten to a hundred times more chunks than they show you, and you pay for every one.