Latent prediction error does not always indicate physical surprise
Framework separates physical surprise from appearance and motion in video models
Stories per day · 30 days
Relatedagents1benchmarks1instruction-following1memory-evaluation1physical-surprise1survey1video-generation1world-models1
8 stories, newest first · one card per story
Stories tagged #video-world-models
Survey unifies video models as reading, writing, sharing, and interaction
The authors treat observed video as a lossy projection of a latent world and track a shift toward persistent state.
MIST finds video world models fail to execute changes, not only memory
M3-Agent more than triples the MIST reveal score by revising instructions so a frozen model executes the change.
WorldMemBench: A Comprehensive Benchmark for Memory of Video World Models
WorldMemBench tests video world-model memory across 260 static and dynamic cases.
HLA-WM: Hybrid Linear Attention for Long-Horizon Video World Models
HLA-WM improves long-horizon video world-model consistency without training while cutting memory use.
Honeycomb: Constant-Size Scene Memory Representation for Video World Models
Honeycomb keeps video world-model scene memory constant using six HexMemory planes.
QuantWM: Temporally Consistent 2-Bit KV Cache Quantization for Video World Models
QuantWM applies training-free 2-bit KV cache quantization to video world models while preserving temporal consistency.
World in World: Explore the World with World Models
World in World is a training-free inference-time interface enabling camera-controlled rerendering, long-horizon revisiting, and motion transfer with frozen causal video world models.