Multimodal AI
Multimodal AI breakthroughs and products
Articles
These startups are chasing the next big thing in LLMs⭐9
Artificial intelligence These startups are chasing the next big thing in LLMs Meet the new kids nipping at the heels of the AI giants.
AI for science needs reasoning, not just data⭐9
Every few decades, someone announces that science has reached its end. In 1903, the revered physicist Albert Michelson wrote that the “facts of physical
UAV3DCrop: Benchmarking 3D Reconstruction in Repeated⭐7
Benchmarking UAV-based 3D crop reconstruction: 88,830 images across 91 scenes reveal trade-offs in appearance, geometry, and scale recovery for agronomic monito...
From Fringe to Policy: How the 'Censorship-Industrial Complex'⭐9
From fringe theory to official policy: How the 'censorship-industrial complex' narrative went mainstream in Washington, shaping legislation and digital governan...
The Download: Censorship Conspiracy Theories and AI-Designed Viruses⭐8
Censorship conspiracy theories move from fringe to Trump policy, plus AI-designed viruses spark biosecurity concerns.
MapTCL: Temporal Consistency Learning via Bidirectional⭐9
MapTCL improves HD map temporal consistency via bidirectional alignment, boosting mAP gains on nuScenes and Argoverse2 without extra overhead.
The Download: Google’s AI Shake-Up and Meta’s Rogue Model⭐9
Meta announces AI shake-up at Google as DeepMind leadership shifts and Jeff Dean exits, plus Meta's rogue AI claims another successful hack.
Advancing Utility Pole and Sign Detection Through Deep Learning⭐10
Deep learning detects utility poles and signs from street imagery, estimating lean angles with 90% mAP via modified DETR, outperforming standard detectors.
The Download: NASA’s New Telescope and Chinese Tech Import Curbs⭐8
NASA’s new telescope can spot killer asteroids while studying dark energy, plus a look at the hidden weakness of metrics.
Hunyuan3D-Buffalo 1.0: A Unified Multimodal Model for Scalable⭐7
Hunyuan3D-Buffalo 1.0 unifies 3D generation, understanding, and editing in a single model, trained on 87M samples, achieving top performance in text-to-3D tasks...