실시간
국내외 뉴스 · 기업 공식 발표 · 논문 · 영상 · 커뮤니티 소식을 한국시간 기준으로 제공합니다. 같은 소식은 하나로 묶고, 기본은 핵심만 보여 드립니다.
모든 수집 소식(구글 뉴스로 들어온 단건 보도 포함).
LLM-based agentic systems are increasingly combined with CoT reasoning to solve tasks that require accurate reasoning, external knowledge, and multi-step…
Prediction errors from video world models are often treated as physical surprise, but they can also reflect appearance, static content, motion, or ordinary…
ODE Policy: Robot Learning via Action Dynamics, Not Actions
대부분의 로봇 정책은 이산적 제어 단계에서 관찰값을 동작값에 매핑하지만, ODE policy는 물리적 운동의 동역학 자체를 학습한다.
From Tool State to Scientific Action: Auditing Decision Provenance in Biomedical Research Agents
BioAgent-DF는 과학 에이전트가 서류 검증 시 근거와 결론의 일치 여부를 평가하고 데이터 이식성을 테스트하는 방법론을 제시한다.
As coding agents achieve increasingly high success rates on software engineering benchmarks, recent efforts have sought to increase task complexity through…
Scientific names increasingly connect papers from different research topics, leaving readers to recover a paper's identity from context.
Coding agents solve repository issues through long sequences of exploration, editing, and validation.
Computer graphics is built on code. Large language model (LLM) agents can now generate graphics code and iteratively refine it through various feedback.
Humans continuously acquire knowledge from carefully crafted materials such as books and tutorial videos, carrying that understanding into future tasks.
Harness optimization can substantially improve LLM agents without modifying model weights, but existing methods primarily optimize response quality.
AI reviewers are increasingly used to assess scientific manuscripts, yet it remains unclear whether their judgments reflect the improvements authors make…
Software development sessions often involve several unfinished tasks rather than one issue carried through to completion.
Long-horizon terminal tasks are used to evaluate and train agents. Human-curated datasets are scarce, so many synthesis methods have emerged.
Heterogeneous embedded boards pair a Linux application processor with a real-time microcontroller (MCU), which suggests moving a task between the two domains…
Anticipating financial outcomes requires synthesizing evolving evidence and reasoning under uncertainty, making deep research systems promising tools for…
Agent decisions depend on empirical relationships that deployment observations reveal and task descriptions alone may leave unresolved.
Streaming video LLMs can now commentate live gameplay, but they are trained and tested on the same games.
Psychotherapy clients may experience distress before they can coherently describe it, and their expression changes with emotion, perceived safety, and trust.
Recent advances in interactive world models have made it possible to generate controllable, dynamic environments from action inputs.
Benchmark compression reduces the often prohibitive time and cost of model evaluation by evaluating models on a small benchmark subset that closely matches…
Fast and Exact Relative-Position Attention via Structure-Aware Kernels
상대 위치 어텐션 메커니즘은 콘텐츠-위치 또는 편향-위치 항목을 추가해 어텐션 점수를 수정하며 오디오 등 분야에서 우수한 성능을 제공한다.
Auriga Technical Report: A Task-Agnostic Agent Harness for Zero-Shot Embodied Manipulation
비전-언어 모델의 고차원 계획과 로봇의 물리적 실행 사이의 격차를 줄이기 위한 작업 조정 에이전트를 제시한다.
Learning to evolve by matching velocities
비선형 역학 계의 연속 시간 진화 학습은 궤적 기반 목표에 의존하지만, 속도 매칭을 통해 이를 개선할 수 있음을 보여준다.
Flide: Accelerating Decision Inference with Exact State Forking
LLM 에이전트 시스템을 위한 의사결정 모델의 추론을 효과적으로 지원하고 가속화하는 엔진을 연구한다.
Multi-head attention in Vision Transformers (ViTs) allows independent attention patterns across heads, but without explicit regularization, heads may learn…
Hierarchical Residuals: A Minimalist Decoupling of Residual Width from Compute Width
기존 잔차 연결은 잔차 폭과 연산 폭을 결합하지만, HiRes 구조는 이 두 치수를 분리한다.
Computer-use agents increasingly tackle tasks ranging from everyday desktop activities to long-horizon professional workflows.
Linear attention scales linearly with sequence length, but its finite-state memory can limit long-context retrieval.
Unlike autoregressive (AR) models, diffusion language models (DLMs) permit parallel token updates, flexible generation, and bidirectional attention, but…
다른 보도 Wes Roth
Molecular discovery requires navigating vast combinatorial search spaces while balancing competing biological objectives.
Training-free expert pruning makes two decisions: which experts to remove, and what the layer computes when its router still selects a removed expert.
LLM-based multi-agent systems (MAS) rely on collaboration among specialized agents to solve complex tasks.
With model weights fixed, we measure how much of an entity-replacement response transfers between English factual questions, beyond whether answers change.
Self-distillation methods for large language models with an exponential moving average (EMA) teacher already retain a history of student updates.
CogBattery: Human-Referenced Evaluation of Cognitive Task Behavior and Structure in Language Models
CogBattery는 13가지 인지 패러다임을 텍스트로 변환하여 54개의 오픈 웨이트 언어 모델을 평가한다.
WMIP: A Working-Memory-Inspired Programmable Probe of Cumulative State Tracking in Large Language Models
대형 언어 모델(LLM)은 중간 단계를 기록하지 않고 단일 프롬프트 내에서 업데이트를 추적하는 과제에 직면해 있다.
Do Multimodal Models Decide Consistently on Equivalent Optical Spectra?
스펙트럼은 나노미터 표나 광자 에너지 플롯으로 제공될 수 있으며 결정은 표현 방식에 의존해서는 안 된다. 답변에 필요한 데이터가 누락되면 모델은 이를 표시해야 한다.
Humanized but Still True? Auditing Facts and Detection in AI-Text Humanizers
AI 휴머나이저는 탐지기가 인간이 작성한 것으로 분류하도록 텍스트를 재작성한다. 공개 리더보드는 회피 및 임베딩 유사성을 평가하지만 검증 가능한 세부 정보의 보존 여부는 평가하지 않는다.
Evidence-Guided Recovery for AI Agents with Ambiguous Tool Outcomes and Partial Execution
도구 시간 초과는 AI 에이전트가 모호한 도구 결과와 부분 실행에 직면하게 할 수 있으며, 이에 대응하는 증거 기반 작업 복구 방식을 정형화한다.