CCTest · Blog
When Do Model Internals Help LLM Safety? A Matched Comparison of DPO and Representation Engineering
AI Safety
cctest.ai
AI Safety

When Do Model Internals Help LLM Safety? A Matched Comparison of DPO and Representation Engineering

A matched study compares DPO, representation steering, internal probes, and text monitors across safety control and risk detection. Representation methods do not replace behavioral alignment, but they offer useful advantages in low-data and cost-sensitive settings.

Read more
CCTest · Blog
VLA-Precision: Making Real-World Online RL More Stable for VLA Robots
Reinforcement Learning
cctest.ai

VLA-Precision: Making Real-World Online RL More Stable for VLA Robots

VLA-Precision addresses policy drift and system overhead in real-world online reinforcement learning for vision-language-action models. Its ACoB algorithm combines staged value calibration with reference-regularized updates, while ACoB-Stream targets the throughput bottleneck of large VLAs.

Read more
CCTest · Blog
TRACE Argues That Streaming Video Models Need More Than One Score
Evaluation & Benchmarks
cctest.ai

TRACE Argues That Streaming Video Models Need More Than One Score

TRACE introduces a condition-aware evaluation framework for streaming video understanding, auditing when evidence becomes valid, how visual history is processed, and when models respond. Its results show that similar QA accuracy can hide major differences in completion, validity, workload, and response timing.

Read more
CCTest · Blog
RayOrch Brings Lineage-Aware Parallelism to Multimodal Data Pipelines
Frameworks & Tools
cctest.ai
Frameworks & Tools

RayOrch Brings Lineage-Aware Parallelism to Multimodal Data Pipelines

RayOrch is a programming model and distributed runtime for data preparation workflows that dynamically split documents and videos into ordered, variable-sized child tasks. It combines cross-parent GPU batching with lineage-preserving result reconstruction and scoped failure handling.

Read more
CCTest · Blog
Teaching Robot Hands to Operate Like Humans: Morphometric Imitation Bridges Retargeting and Zero-Shot Deployment
Robotics & Physical AI
cctest.ai

Teaching Robot Hands to Operate Like Humans: Morphometric Imitation Bridges Retargeting and Zero-Shot Deployment

Researchers from UC Berkeley and collaborators introduce Morphometric Imitation, a three-stage pipeline that turns reconstructed human hand-object interactions into executable robot demonstrations and visuomotor policies. Across three robot hands and ten interaction tasks, the method reached a 89.3% zero-shot success rate in real-world trials.

Read more