AI House law

The training data is the boundary

A model can only act on distinctions it was shown. What never appeared in training does not register as a hard case at all, so the output stays confident and wrong in the same places every time.

No pieces have proved this law yet. The first one is on its way.

Privacy preferences

Choose which cookies we can use. You can change this anytime from the footer.

Strictly necessary

Keeps you signed in and remembers these choices. The site needs these, so they stay on.

Analytics

Helps us see which stories land so we can write better ones. Our current analytics is cookieless.

Advertising

Lets Google show ads that keep the site free. Turn this off to see non-personalised ads instead.

Newsletter

Unsubscribe anytime · Privacy