Import AI 476 examines creativity and recursive self-improvement, open-weight models that perform worse than China's, and AI in space.

Meta Description: Elorian and Scale AI released Humanity’s Sixth Sense, a 522-question benchmark testing implicit visual reasoning across images, video, and scene logic.

Tags: AI benchmarks, visual reasoning, Elorian, Scale AI, Humanity’s Sixth Sense

Featured Image: Abstract technology-style visual reasoning grid with implied bookcase, plough, and chain motifs, no people, no logos, no text.

AI startups Elorian and Scale AI have built and released Humanity’s Sixth Sense (HSS), a benchmark that measures how a model interprets and reasons through information a visual scene implies but does not explicitly show. The benchmark arrives as AI systems become more fluent at naming objects, but still struggle with the unstated physics, social cues, and causal patterns that people infer from a single image or short video.

Background: a gap that captions can miss

Current AI models can often describe visible elements in a photo, yet the harder question is what the scene suggests. HSS targets that gap. In one example, a picture of a bookcase asks whether it can fit more books, requiring spatial judgment rather than simple object recognition. In another, a photo of horses pulling a plough asks whether the chains bear the plough’s drag weight, requiring physical reasoning about force and load.

These cases show why a new visual reasoning benchmark is useful. The goal is not to score a model on labels, but on whether it can read implied structure: whether something will fit, whether a connection can hold, and whether an observer can infer what is not shown.

Development: 522 questions across images and video

The released benchmark contains 522 questions that cover 288 images and 234 short video clips, totalling 17.6 hours of video. The questions fall into four categories.

Temporal & causal dynamics asks what happened before the scene and what happens next. Physical & spatial logic asks what fits, what is in reach, and what someone else can see from where they are standing. Social understanding focuses on intent, attention, and belonging. Abstract & contextual inference looks for the patterns and rules a scene implies.

This structure gives HSS a timeline-like shape: it starts with visible evidence, moves through motion and physical constraints, then adds people and hidden rules. For readers, that means a model’s score on HSS is less about image recognition and more about whether it can reason across time, space, society, and abstraction.

Latest progress: comparison scores in the release

The release material also lists comparison scores. DeepSWE 1.1: 44.4, versus 61 for GLM-5.3. GPQA Diamond: 90.5, versus 91.7 for GLM-5.3. MCP Atlas: 78.7, versus 84.2 for GLM-5.3. Artificial Analysis - DeepSWE 1.1: 62, versus 68 for Kimi K3 and 61 for GLM-5.3. Terminal-Bench 4.0: 28.3, versus 41.9 for GLM-5.3. Finance Agent V2: 54.7, versus 55.8 for GLM-5.3.

These numbers sit alongside the HSS release and show how AI evaluations are being compared across different tests, including GLM-5.3, Kimi K3, and Artificial Analysis results. The pattern is useful for readers because it separates broad capability scores from visual inference: a model can perform strongly on some benchmarks while still failing to interpret a scene’s hidden physical or social logic.

Next phase: spacecraft scenarios and first-mover questions

The release material also expands beyond screens into spacecraft and AI systems. Spacecraft can operate autonomously. The argument is that designers can redesign spacecraft around the assumption they will be hosting AIs rather than humans. A speck of space debris that could puncture the International Space Station (ISS) with fatal consequences might leave an equivalent but uncrewed, unpressurized vessel largely functional.

The material also notes that space-based assembly and later manufacturing will let AI build large, complex things. It then asks whether first movers could gain compounding advantages. If an AI civilization built up a foothold in space, it might become economically and strategically dominant. One example is an industrial flywheel where initial infrastructure begets more energy and manufacturing capacity at lower costs, which begets more infrastructure. The material also notes space’s great domain awareness, where it may be possible for a space-based power to deny access to space for others, and first movers could send self-replicating probes into deep space.

At this stage, HSS is a new visual reasoning benchmark with 522 questions, while the spacecraft discussion points to a longer-term question: whether AI systems can move from interpreting scenes on Earth to operating and building in environments where human presence is no longer assumed.