Maligned - December 13, 2025

AI news without the BS

Here’s what actually matters in AI today. No fluff, no hype - just 5 developments worth your time.

Today’s Top 5 AI Developments

1. Robots Feel the Force: Vision & Touch Come Together 🤖

Researchers just dropped ImplicitRDP, a new diffusion policy that finally bridges the gap between slow visual planning and rapid force sensing in robotics. It allows robots to perform complex, contact-rich tasks by processing both modalities within a single network, leading to better reactivity and higher success rates. This means robots can now “feel” their way through tasks with human-like dexterity.

Source: arXiv Link: https://arxiv.org/abs/2512.10946v1

A novel approach called Group Diffusion lets images collaborate during inference, unlocking cross-sample attention that significantly boosts generation quality. Instead of each image being denoised independently, they learn together, resulting in up to 32.2% FID improvement on ImageNet. This fundamentally changes how diffusion models can learn and generate higher quality outputs.

Source: arXiv Link: https://arxiv.org/abs/2512.10954v1

3. Faster, Better Generative Flows: A New Spin on Data-to-Noise 💨

Introducing BiFlow, a Bidirectional Normalizing Flow that ditches the need for an exact analytic inverse, making generative flows more flexible and efficient. It approximates the inverse mapping, leading to improved generation quality and sampling speeds up to two orders of magnitude faster. This is a big step forward for a classical generative modeling paradigm.

Source: arXiv Link: https://arxiv.org/abs/2512.10953v1

4. Autonomous Cars Get Smarter Eyes: Stereo Vision & Mid-Level Cues Win Big 🛣️

StereoWalker is challenging the “end-to-end” robot navigation foundation model hype by proving that explicit stereo inputs and mid-level vision (like depth and tracking) are still critical for robust navigation. It achieves state-of-the-art driving performance with a tiny fraction (1.5%) of the data compared to monocular-only models, showing that some “implicit” assumptions just aren’t cutting it in dynamic urban scenes.

Source: arXiv Link: https://arxiv.org/abs/2512.10956v1

5. Pinpoint Control for AI Art: Isolate Any Attribute You Want 🎨

Omni-Attribute is a game-changer for visual concept personalization, letting you transfer specific image attributes like identity or style without leakage. By jointly designing data and model to explicitly teach the encoder what to preserve or suppress, it provides unprecedented open-vocabulary control, enabling high-fidelity, attribute-specific representations for generation and editing. No more messy attribute entanglement.

Source: arXiv Link: https://arxiv.org/abs/2512.10955v1

That’s it for today. Stay aligned. 🎯

Maligned - AI news without the BS