Import AI 476: Creativity and RSI; worse-than-China open weight models; AI goes to space
Will we give robots the moon?
Welcome to Import AI, a newsletter about AI research. Humanity's Sixth Sense is a new benchmark that tests how well AI models can interpret and reason through visual scenes that don't explicitly provide all the information. The benchmark contains 522 questions covering 288 images and 234 short video clips (17.6 hours of video in total).
It assesses four categories: temporal and causal dynamics, physical and spatial logic, social understanding, and abstract and contextual inference. Human performance on this benchmark is quite high, with an aggregate pass@1 score of 93.1, but AI systems like GPT 6 Astra only reach a pass@1 score of 53.6. Notably, Chinese open weight models perform even worse, with scores of 26.8 for GLM-5.3-flash and 25.5 for Kimi K3.
This benchmark reveals that current AI models lack intuitive visual reasoning, suggesting that such reasoning may not necessarily require specialized world models. Meanwhile, AI startups Elorian and Scale AI have developed and released Humanity's Sixth Sense as a benchmark to measure how well models interpret and reason through visual information.
The AI race is also heading towards space, where AI systems could accelerate the industrialization of the solar system. Spacecraft can operate autonomously, and with the advent of Starship, the space economy may boom, enabling AI to build large, complex structures. This could potentially create a compounding advantage for AI civilizations that first establish a foothold in space, allowing them to dominate economically and strategically.
Written by urgent.news from Import AI's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
