An experimental learning project by Loophole
Learning, one example at a time.
JAYCE explores how labeled examples and corrections can improve future guesses using Adaptive Prototype Memory. Try the lesson below, then explore the code.
JAYCE is available under the GNU Affero General Public License v3 (AGPL-3.0-only).
JAYCE · ADAPTIVE PROTOTYPE MEMORY
Watch Jayce learn toy names
Inspired by toddler Jayce learning toy names through examples and corrections, JAYCE explores how labeled experiences can improve future guesses. This lesson shows the project's Adaptive Prototype Memory (APM) method.
Jayce's toy bins
6 labels · 2 memory slots eachEarlier memories appear first; new toy types fill the later bins. An exemplar is a specific example kept for comparison. Each prototype starts with one exemplar's numbers; later updates can blend in other examples. The computer does not see or understand the toy itself.
How Jayce compares examples
Toy features, simplifiedWhy are toys close?
We chose two illustrative scores from 0 to 1: body length and body bulk (solid bodywork). Nearby toys have similar scores, even across labels. A bike has two wheels; a skateboard and quad roller skate have four. Wheel count alone cannot separate the latter two. Other features could help and would change the map.
What does the correction do?
In these examples, a person supplies the correct label and Jayce saves the toy in that label's empty slot. The earlier Truck or Car prototype stays where it was. The correction adds a correctly labeled example; it does not relabel an older one.
Check the distances
Square each difference, then add themUse the example cards to see why a guess wins and how a correction changes the next guess.
Use Next and Back at your own pace. Play advances every 8 seconds; Pause stops it.
JAYCE means “Jayce Associates Your Categorized Exemplars.” APM compares number lists and uses labels supplied by a person.
Copyright © 2026 Loophole, LLC. AGPL-3.0-only · No warranty. Source code.