Gemini in Reasoning: Unveiling Commonsense in Multimodal Large Language Models

Meta info.

• Authors: Yuqing Wang, Yun Zhao
• Paper: https://arxiv.org/pdf/2312.17661.pdf
• Affiliation: Meta AI, Stanford Univ.

TL; DR

Gemini ≒ GPT-3.5-turbo, Gemini ≲ GPT-4-Turbo. Gemini는 temporal, social reasoning과 이미지에서 감정을 읽는 task에 약하다고.