Illustrative

Small Model Forum

When is a small model enough?

  • FormatOne day, in person
  • CityTo be confirmed
  • Date and venueTo be confirmed
  • You leave withA test for your own data

Session plan

A small model's decision is only as good as the point where you would have asked for help.

  1. A confident, wrong answer.

    One author brings a case where a cheap model was confident and wrong, and walks through the prediction at the threshold for calling a larger model.

  2. One piece of evidence removed.

    A second researcher removes a retrieved passage from the same case.

  3. The first point where the decision changes.

    The room looks for the first change that flips the routing decision, and argues about why.

  4. A test to take home.

    Participants leave with a test they can try on their own data.

What to bring

  • A case from your own work where a cheap model was confident and wrong.
  • A rough idea of what you could have measured before sending the answer.
  • A laptop, if you want to try the test during the working session.

Would you go to a meeting like this?

Tell us the city and the topic. This page is an illustration, and no meeting is scheduled.