MERIT is a new AI framework that learns to separate melody, rhythm, and timbre components of music similarity. Unlike existing tools that treat a song as a single fingerprint, it lets you ask: do these two tracks share the same groove?
PARE achieves 52% parameter reduction on Wan2.1-14B with just 0.6 points drop on VBench, using spatial-temporal aware pruning and content-adaptive routing. Combined with step distillation, total speedup reaches 50x.
A new technique from Linköping University trains diffusion models on DINO self-supervised representations instead of text prompts. The model discovers semantic variation directions in image data without any labels.
Security researchers have found a way to hijack voice AI models using inaudible sounds embedded in ordinary audio clips. AudioHijack achieved 79 to 96 percent success rates across 13 large audio language models.
arXiv is now banning authors for one year after detecting AI-hallucinated citations in submitted papers. With fabricated references appearing in 1 in every 277 papers, here is what every creator using AI for research needs to know.
Richard Socher's stealth-mode AI lab Recursive exits with $650M at a $4.65B valuation, backed by GV, Greycroft, Nvidia, and AMD.