Machine learning news — deep learning, reinforcement learning, neural architecture search, diffusion models, and new ML frameworks and libraries.
Researchers have systematically defined various notions of AI-generated text and found that current detectors often perform well only for specific definitions but fail as broad detectors. The team introduced AITDNA, a ne…
Researchers trained a small GPT-2 model on arithmetic problems using an Indonesian pedagogy called GASING, which breaks down calculations into left-to-right steps aligned with token generation. The 86-million-parameter m…
Researchers have developed a method to automatically generate research paper titles from abstracts using open-weight pre-trained and large language models, including fine-tuned PEGASUS-large and zero-shot GPT-3.5-turbo. …
Researchers have identified a deployment-safety gap in vision-language-action (VLA) robot policies, showing that identical model checkpoints can produce different physical actions due to variations in action unnormalizat…
Researchers have developed FORGE, a multi-agent system that bridges vulnerability exploitation, prioritization, and detection rule engineering by using graduated exploitation depth. The system achieved 67.8% end-to-end e…
Nvidia has acquired Kumo AI, a four-year-old startup that builds predictive foundation models for business data, for at least $400 million according to The Information. The three Kumo co-founders moved to Nvidia last mon…
A new open-source benchmark, "apple-silicon-llm-bench," reveals that Google's LiteRT-LM runtime outperforms MLX-Swift on the iPhone 17 Pro for Gemma 4 E2B inference, achieving 55.4 tok/s with 4.5x less memory usage. The …
A new paper situates large language models within the longer history of computational approaches to analyzing scientific concepts in the history, philosophy, and sociology of science. The study reconstructs computational…
Researchers have identified a new failure mode in multi-step agentic retrieval-augmented generation (RAG) systems called cascading hallucination, where errors from early pipeline stages propagate and amplify across reaso…
Researchers have developed IN2R, a new framework that corrects mismatched image-text pairs in large web-harvested datasets by synthesizing continuous supervision signals from intra-modal data relationships rather than re…
Researchers proposed Answer Self-Consistency with Margin-Triggered Question Re-Arbitration (ASC-MQRA) for the CVPR 2026 VidLLMs Challenge Track 2, a training-free test-time reasoning framework that improves video relatio…
Researchers have extended refusal steering methods to Mixture-of-Experts (MoE) large language models, demonstrating that steering vectors can effectively suppress safety-aligned refusal behavior in these architectures. T…
Researchers have introduced DLLG (Dynamic Logit-Level Gating), a framework that learns token-level fusion weights for combining multiple specialized large language models without requiring token-level labels or expert re…
Researchers have developed a differentiable Neural Architecture Search (NAS) framework that jointly optimizes both architectural configurations and mixed-precision quantization for compressing large language models (LLMs…
Researchers introduced SDPG, a self-distilled policy-gradient framework that combines group-relative verifier advantages with normalized standard deviation and full-vocabulary on-policy self-distillation to improve reinf…
Researchers have introduced Trivium, a causal-memory controller that treats temporal regret—how long a miscalibrated causal model persists—as a first-class objective alongside outcome and epistemic regret. The system add…
Researchers have developed LazyAttention, a new attention mechanism that enables zero-copy, position-agnostic key-value cache reuse for large language models by deferring positional encoding to within attention kernels. …
Researchers introduced a methodology using CausalBERT, enhanced with temperature scaling, hyperparameter optimization, and interpretability methods, to isolate the impact of specific factors on online review ratings. The…
A large-scale study of biomedical question-answering systems found that retrieval-augmented generation (RAG) provides only small and inconsistent accuracy improvements—typically 1-2 percentage points—over models that do …
Researchers have developed RUBAS, a rubric-based reinforcement learning framework designed to improve safety in large language model (LLM) agents that execute real-world tasks. The framework evaluates agent behavior acro…