When testing the last3+text_summary settings we noticed an OR effect compared to either last3 or text_summary alone for Kimi. This meant that for some the images helped the model more and for others the text helped, and except for one maze that text summary could solve alone. We should look into this further for an exploration of what parts of this augmentation work
When testing the last3+text_summary settings we noticed an OR effect compared to either last3 or text_summary alone for Kimi. This meant that for some the images helped the model more and for others the text helped, and except for one maze that text summary could solve alone. We should look into this further for an exploration of what parts of this augmentation work