CAN YOU INFINITELY LEARN WITH ONLINE RL?

Summary from the TLDR AI newsletter

Geospot Infinity is a photo-to-GPS model. It retrieves 10 candidate coordinates for each upload and ranks them for the user. Users then pick the closest. The model attempts to learn from interactions via online reinforcement learning. It turned out that over 65% of users would just click the first guess regardless of accuracy, effectively nullifying the model's ability to learn anything.