Northeastern's CATT Benchmark Exposes Why AI Data Attribution Misses Its Sources
A new benchmark called CATT shows why data attribution methods fail, and two simple fixes boost recall by 35 to 72 points.
20 stories
A new benchmark called CATT shows why data attribution methods fail, and two simple fixes boost recall by 35 to 72 points.
Kyutai researchers show that a random bias on the output logits reproduces most gains attributed to Neural Thickets weight perturbations, with a fully frozen model.
KAIST's L3G-World pairs a video diffusion transformer with a 3D Gaussian memory that stores VAE latents, hitting 9 FPS interactive world generation.
A CMU-led team collapses pretraining, SFT, and RL into one maximum-likelihood objective, letting you mix static text and RL rollouts in a single trainer.
Prime Intellect added a backward pass to DeepSeek's forward-only Mega MoE kernel, cutting Qwen3 training step times by up to 55%.
Pika releases an open-weight video model post-trained on MiniMax H3 that generates 15-second clips with audio in 6 denoising steps.
Anthropic ships multi-account switching, CarPlay voice mode, 24 new languages, and a laptop-style Android layout for Chromebooks and foldables.
Jev and a local model matched my reference on 12 of 12 Claude Code fixes to a synthetic API. A filename rule matched 3 and still caught the one real bug!
NVIDIA's LucidLyra turns one image into a 3D Gaussian scene you can query by text, with semantics and instances baked in from a single feed-forward pass.
IFM has released K2 Horizon 360, bundling 380 checkpoints, 13T training tokens, RL infrastructure and a new attention architecture under Apache 2.0.
Gemini's file generation feature now includes a full in-chat editor for documents, PDFs, slides and spreadsheets, with highlight-to-edit prompts.
Phonic 1 (High) debuts at the top of Artificial Analysis's Speech to Speech Index at 87.1, leading agentic voice tasks while running at $0.07 per minute.
A 2.6GB uncensored local model with 1M-token context, thinking mode, and tool calling that runs on laptops or even phones without internet.
Cohere closes initial tranche of a CAD $7 billion Series E at a CAD $35.7 billion valuation, the biggest private round in Canadian history, to challenge US and Chinese frontier labs.
A new distillation method shows flow-based robot policies behave like Gaussians, cutting inference time in half with near-identical task success.
Fish Audio's new 35B mixture-of-experts speech model takes the top open-weights spot on both Artificial Analysis arenas and leads all models on pronunciation.
Mistral released the open weights for its 1-trillion parameter Mistral Large 4 model, now leading open-weight coding benchmarks with multimodal input and cyberdefense tuning.
PhysAg3D turns a single photo into a rigged, physically interactive 3D asset by pairing a mesh generator with a vision-language agent that reasons about joints and deformation.
A tiny U-Net learns to turn chest vibrations into 12-lead ECG waveforms, with a dataset that finally includes precordial lead positions.
Poolside's 225B Mixture-of-Experts coding model lands with open weights, 262K context, and benchmark scores rivaling frontier closed systems.