News

Researchers developed fMRI-Diffusion, a framework that generates synthetic fMRI time series using a Temporal Transformer diffusion model to improve Major Depressive Disorder diagnosis. The method outperformed existing au…

Raon-Speech, a 9-billion-parameter speech language model for English and Korean, and its full-duplex extension Raon-SpeechChat achieve top performance across 42 speech and text benchmarks, surpassing similarly sized mode…

Researchers have developed the Multi-Persona Debate System (MPDS), a framework that uses large language models to automatically generate scientific hypotheses by simulating structured debates between literature-grounded …

Researchers have developed EchoDistill, a self-distillation framework that improves Audio Large Language Models' resistance to real-world noise by aligning noisy student responses with clean semantic references from a fr…

Researchers at arXiv introduced a causal framework to detect rationalization bias in LLM judges, finding that these models often fabricate explanations to justify rankings influenced by non-evidential cues like verbosity…

Researchers introduced AERIC, a lightweight safety monitor that detects implicit harmful dialogue by reading a language model's internal hidden states during ordinary text generation, requiring only 387 trainable paramet…

Researchers have identified three systematic failure modes in Audio LLMs when transcribing English-Mandarin code-switching speech: language omission, translation-instead-of-transcription, and hallucination. By applying D…

A multi-probe audit of clinical-interview depression detection benchmarks found that a lightweight hybrid text-plus-LLM-score model achieved a macro-F1 of 0.723 on the E-DAIC dataset under strict subject-disjoint cross-v…

A new analysis of approximately 20 million chat messages from 4,452 Twitch streams across seven game genres found that 2.4% of all messages were toxic, with MOBA games showing the highest rate at 3.2% and sports games th…

Researchers at arXiv have introduced "infilling extraction," a new method for extracting training data from diffusion language models (DLMs) that uses arbitrary binary masks instead of relying solely on prefix-conditione…

Researchers have developed a modular pipeline for generating educational analogies, breaking the task into four stages: source finding, sub-concept generation, explanation generation, and evaluation. Testing 12 large lan…

Researchers released QUEST, a family of open-source AI models ranging from 2B to 35B parameters, designed to function as general-purpose deep research agents for long-horizon search tasks. The models, trained using a ful…

A new AI-driven workflow that uses detailed "constitutions" to define labeling categories and a frontier LLM to interpret them has reduced cross-model inconsistency by up to 57 times compared to standard paragraph defini…

Researchers have developed SteER, a framework for steerable deep research that allows users to provide mid-process input during long-form research tasks. The system uses cost-benefit analysis to decide when to pause for …

Researchers introduced DRinQ, a benchmark designed to evaluate how well large language models understand conversational implicature—meaning suggested rather than explicitly stated—by systematically varying context while …

Meta released CacheLib 2026.05.25 on Monday, its first new version in two years, to help services cope with surging DRAM prices driven by the AI boom. The open-source caching engine, originally launched in 2021 to offset…

A developer built a self-hostable, web-based SQL client for teams, offering features like SSO, audit logs, and a modern editor, to address security and functionality gaps in existing tools like DBeaver and PopSQL. The to…

A developer built a new programming language, Jda, that compiles 33 times faster than Rust, achieving an average compile time of 42 milliseconds. The single-pass compiler produces 1MB static binaries and includes built-i…

A developer expanded their AI-agent security benchmark from 10 to 16 scenarios, revealing that Claude Code Sonnet 4.6 scores +9 out of 16 while Haiku 4.5 scores only +3. The original tie between the two models was a smal…

Caution Labs has built AI-powered content moderation infrastructure designed to detect Personally Identifiable Information (PII) across text, images, and AI-generated workflows. The company warns that PII—including names…