Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
Researchers evaluated multimodal large language models (MLLMs) for assistive AI applications, finding they show promise in object recognition and multilingual text reading but have limitations in real-world egocentric ta…
Researchers at arXiv found that in language models, the direction that best detects a behavior and the one that best controls it are often geometrically distinct, with cosine similarities as low as 0.12 for hallucination…
A new survey analyzes the use of large language models (LLMs) for scientific peer review, focusing on critique generation and score prediction. The study identifies reliability, robustness, and security risks such as pro…
Researchers at arXiv found that jailbreak attacks on large language models can be detected by analyzing entropy dynamics in intermediate layers, rather than final outputs. The study shows that monotonic rank-based trend …
Researchers propose G-SPIN, a graph-based framework that corrects phonetic errors in ASR outputs by combining a graph neural network for candidate generation with a masked language model and LLM for contextual re-ranking…
Researchers introduced iLLaDA, an 8B masked diffusion language model trained from scratch with fully bidirectional attention, scaling pre-training to 12T tokens and fine-tuning on a 25B-token instruction corpus. Compared…
Researchers introduced Hybrid-IR, a dual-path retrieval framework with iterative reasoning for complex medical question answering. The system combines graph-based and dense retrieval to overcome limitations of single-pat…
A small group of Wikipedia editors, the Pro-Animal Wikipedians (PAW), measurably shaped how large language models discuss animal welfare through just 125 edits across 115 pages. Gradient-based data attribution and counte…
Researchers introduced Local Branch Routing (LBR), a token-level test-time scaling framework that expands a small local lookahead tree and uses a lightweight router to select the best branch, enabling efficient and train…
Researchers developed MRI2Rep, an autoregressive framework that generates structured reports for 3D liver MRI, achieving 76% case-level sensitivity and 82.4% liver-level accuracy on a test set of 3,929 real-world MRI-rep…
Researchers propose Dustin, a sparse verification framework for long-context speculative decoding in LLMs, achieving a 27.85x speedup in self-attention and 9.17x end-to-end decoding speedup at 32k sequence length with ne…
Researchers introduced a dataset of 32,534 double-marked GCSE student responses and tested large language models (LLMs) against examiner consensus. Top-performing LLMs agreed more closely with examiners than examiners ag…
Researchers propose SALSA, a single-pass autoregressive LLM structured classification method, for detecting machine-generated code in SemEval-2026 Task 13. Their system achieves an out-of-distribution F1 score of 0.789, …
Researchers propose a test-time adaptation method using semi-supervised learning to improve AI text detection under continual distribution shifts, such as adversarial humanization and new LLMs. The approach leverages inf…
Researchers at arXiv have developed a prompt-based learning method for automatically generating highlights of academic papers, using language models like ChatGPT. The approach achieves performance comparable to supervise…
Researchers present a low-resource machine translation system for the Tangkhul-English language pair, achieving a BLEU score of 39.97 using a ByT5-large model fine-tuned on 38,336 parallel sentences. The study highlights…
A new survey from arXiv redefines continual learning for large language models as a closed-loop update-and-release problem within a versioned ecosystem, identifying three core challenges and five lifecycle design princip…
Researchers propose FreeStory, a training-free framework for visual storytelling that maintains character consistency across images under free-form narrative prompts, where characters are introduced once and later referr…
Researchers released MacroLens, a multi-task benchmark for contextual financial reasoning under macroeconomic scenarios, covering 4,416 U.S. small- and micro-cap equities from 2021-2026. The benchmark integrates prices, …
Researchers propose an empirical test for error correction in large language models, finding that residual-stream activations are robust to small perturbations and that feature-specific directions are privileged over gen…