Korva is a scroll of the internet’s best ideas.

Picture a coconut water brand with endless product shots generated inside the ad platform: the same bottle with New York scenery behind it for one viewer and California scenery for another.

Meta has been pushing a goal-only ad machine, where an advertiser states a goal such as reaching a certain number of people, hitting a view count, and getting them to buy a product. The system does the rest of the work behind the scenes, and an image model supplies the assets.

Because Meta sees the full purchasing loop, it can tell that one AI-generated image converted better than another and run live A/B tests on different audiences. Cheaper ads mean brands buy more of them, which funds better image models.

How does an A/B test pick a winning ad?

The platform randomly splits an audience into groups, shows each group a different version, and compares the rate of a chosen action such as purchases. Random assignment means differences in outcomes can be credited to the ad and not to the audience. A statistical test then estimates whether the gap is bigger than chance alone would produce. The same logic underlies clinical trials, which trace to James Lind's 1747 scurvy experiment on twelve sailors.

Why do cheaper ads lead to more ads?

When a cost falls far enough, demand can grow by more than the cost shrinks, so total spending rises. In 1865, economist William Stanley Jevons observed that more efficient steam engines increased coal use in Britain instead of reducing it. The observation is now called the Jevons paradox. He made it in his book The Coal Question.

How does an ad auction choose who wins?

Meta describes its ad auction as ranking ads by total value, not by bid alone. Total value combines the advertiser's bid, the estimated chance that the viewer takes the desired action, and a measure of ad quality. The ad with the highest total value wins the slot. Advertisers can therefore win with a lower bid if their ad is more likely to get the action.

Enjoyed it?