Christopher Ort
-
Teaching LVLMs Spatial Skills With 15K Toy Blocks Beats Heavier Real-Scene Pipelines
SpatialBlock-15k: Qwen-based 4B model reaches 51.3% (+25.1 over backbone). 7B gains 17.6% over backbone; 3B beats SpatialLadder by 2.7–3.2% using only synthetic blocks.
-
TurboVLA Runs Robot Policies at 32 Hz on an RTX 4090 Using Under 1 GB VRAM
HUST’s 0.2B TurboVLA hits 97.6% LIBERO success at 31.2 ms latency and 0.9 GiB VRAM — ~8× faster than π0.5 with ~6% of the parameters.
-
YuE2’s Symbolic Planning Wins 49.3% of Expert Preferences for Music Quality
YuE2 writes a readable score, then audio. Same-checkpoint experts prefer planning 49.3% vs 34.6% without. WildSongBench Global Avg 6.73 (6.96 best-of-8), topping public baselines.
-
VLMs Still Can’t Tell Which Attribute Goes With Which Object — Until Text-Only Finetuning
SABER binding benchmark: an 8B model finetuned on synthetic text alone jumps from 52.5% to 84.2% mean accuracy, with +34.1 points text–image — no visual training…
-
Qwen-Image-2.0 Unifies High-Fidelity Generation and Precise Editing in One Model
Alibaba’s Qwen-Image-2.0 pairs Qwen3-VL with a Multimodal DiT and a 16× VAE. It targets long text, multilingual typography, high-res photorealism and editing without pipeline switching.
-
Crisis Deepfake Detectors Collapse From 90%+ AUC to Chance on Real Generators
RA-Bench: traditional detectors fall from 67.6–98.6% public AUCs to 43.9–57.3% source-level means. Social dissemination cuts fine-tuned FakeR from 46.0% to 1.4%.
-
Latent Visual Reasoning Barely Notices When the Image Changes — ReaLVR Fixes That
Amazon/UAB study: LVR models change predictions only 5.7–13.1% when image edits flip the answer 81–86% of the time. ReaLVR hits 63.7% five-task average on Qwen2.5-VL-7B.
-
Vidu S2 Brings Real-Time 720p Interactive Avatars and Live Video Editing
Tsinghua/Shengshu’s Vidu S2: real-time 720p avatars with dynamic references, plus streaming edits (style, try-on, character/background swap). Human GSB prefers it 85.7–100% on overall quality vs top…
-
Chunked KV Compression Creates Periodic Blind Spots Up to 40 Points Wide
DeepSeek-V4 family retrieval accuracy swings by as much as 40 pp depending on a token’s phase inside the compression window — gaps average scores hide.
-
Diffusion-Augmented LLMs Deliver Up to 3× Lossless Speedups Over Autoregressive Decoding
Uno adds lightweight diffusion weights to an AR LLM for parallel multi-token draws. It beats speculative decoding throughput at every batch size and speeds RL training…