Hai-Wen Diao

GPT-6-Astra achieves unprecedented embodied intelligence performance, yet practical deployment faces significant hurdles in demonstration cost and physical constraints.

Toy games are surprisingly good proxies for real-world visual reasoning, but training on them can degrade fine-grained perception.

A 260M-parameter looped diffusion model achieves performance surpassing a 6.5x larger non-looped counterpart with significantly less inference compute.

Unified models can master multi-turn visual self-correction through end-to-end trajectory RL, eliminating the need for external critic models or inference-time verifiers while yielding double-digit visual alignment gains.

VA-Bench is introduced to evaluate the complete observe-reason-act-revise loop and tests whether general-purpose MLLMs can turn visual demonstrations and actively acquired evidence into successful embodied action.

Native unified modelling is position as a promising path towards systems that perceive, reason and create within a fully end-to-end framework through SenseNova-U1.5, an 8B-MoT native unified multimodal model that understands, reasons about, and generates visual content within an encoder-free and VAE-free architecture.

Task-decoupled architectures can unlock the full potential of unified multimodal models, transforming coexistence into synergy for complex tasks.

Rubric-based scoring reveals that vision-language models can significantly improve their grounding in visual evidence, enhancing reasoning and instruction adherence.

Models trained with VBVR-Pro not only excel in a controlled task space but also show significant transferability to external benchmarks, revealing critical insights into visual reasoning mechanisms.

A single unified model can outperform specialized systems across various computer vision tasks, all without the need for custom architectures.