Papers

“Hy-Nomas is a video world model that uses causal attention and structured conditioning to generate interactive, multi-turn vide…”

Published 2026-10-26

Fetched 2026-10-28 23:01 UTC · ed5f2c71ab05e085

Torn from — the abstract

“Multimodal language models (MLLMs) exhibit promising spatial reasoning capabilities, yet the internal mechanisms supporting the…”

Published 2026-10-09

Fetched 2026-10-12 23:01 UTC · dd7bd54bf7643623

Torn from — the abstract

“Coding agents are increasingly used to build programs from scratch as software engineers do rather than to resolve issues withi…”

They claim — Fine-tuning on trajectories with these constructed environments raises the performance of Qwen3.8-27B from 34.87% to 42.84%.

Published 2026-10-08

Fetched 2026-10-12 23:01 UTC · 221fc2bb32cab742

Torn from — the abstract

“Continual learning requires models to accumulate past experience and transfer it to new tasks. Despite substantial research on …”

Published 2026-10-08

Fetched 2026-10-12 23:01 UTC · 1725bcbca9c55585

Torn from — the abstract

“Continual post-training faces a dual challenge of acquisition and retention: models must learn new capabilities while preservin…”

They claim — Evaluations spanning 4 acquisition datasets (including a mixture) and 14 retention benchmarks show SynthSelf consistently outperforms 6 methods (including off- and on-policy distillation)…

Published 2026-10-08

Fetched 2026-10-12 23:01 UTC · ecac050336939f35

Torn from — the abstract

“Computer-use agents can perform increasingly complex GUI tasks, yet unfamiliar software and workflows requiring human domain kn…”

They claim — When the same four execution models are equipped with skills induced by each method under shared execution settings, Work2Skill achieves the highest scores and success rates.

Published 2026-10-08

Fetched 2026-10-12 23:01 UTC · 0bf731e1245f4493

Torn from — the abstract

“A 4D audio-visual scene comprises a video, the dynamic geometry it depicts, and the sound sources that populate it, each with i…”

Published 2026-10-07

Fetched 2026-10-12 23:01 UTC · f8acf62ebace4a63

Torn from — the abstract

“Online reinforcement learning (RL) fine-tunes diffusion models with black-box rewards, but reverse-process methods such as Flow…”

They claim — In particular, on HunyuanVideo-13B with equally weighted visual and motion quality rewards, DiffusionART reaches the late-stage mean reward levels of DiffusionNFT and Flow-GRPO-Fast with …

Published 2026-10-07

Fetched 2026-10-12 23:01 UTC · 789a779d68f1191e

Torn from — the abstract

“Humans effortlessly react to the world; they adeptly return a tennis shot, catch a falling cup, step aside from a passing cart,…”

They claim — In simulation, Reflex achieves 85.7% catch success from egocentric RGB-D, compared with 87.9% when given privileged box observations.

Published 2026-10-07

Fetched 2026-10-12 23:01 UTC · a3843271a70bb133

Torn from — the abstract

“Computer-use agents (CUAs) are rapidly improving, yet synthesizing diverse and reliable software environments at scale remains …”

They claim — Using SoftwareGym, we synthesize 182 applications across twelve categories and achieve a mean GUI task pass rate of 91.5%.

Published 2026-10-07

Fetched 2026-10-12 23:01 UTC · 5a26c82bd2ed4195

Torn from — the abstract

“Multimodal large language models have become the dominant paradigm for visual understanding, but incur substantial costs by enc…”

They claim — Based on Qwen3.8-27B, VisionWeave adaptively adjusts token savings to visual content, saving 43.0% tokens on average while retaining 98.9% native performance across eight benchmarks, vers…

Published 2026-10-05

Fetched 2026-10-12 23:01 UTC · 57525ad9734267df

Torn from — the abstract

“Diffusion large language models dLLMs) have emerged as a promising alternative to autoregressive language models through bidire…”

They claim — Across full-sequence dLLMs (LLaDA and Dream) and block-diffusion dLLMs(Fast-dLLM v2 and Nemotron-Labs-Diffusion), SoloQ retains accuracy under 4-bit quantization and outperforms calibrati…

Published 2026-10-04

Fetched 2026-10-12 23:01 UTC · 2ba5dc712e71f81e

Torn from — the abstract

“Recent joint video-audio generation models have achieved strong semantic correspondence and temporal synchronization. However, …”

Published 2026-09-29

Fetched 2026-10-12 23:01 UTC · 7768b360a8a3a4b5

Torn from — the abstract

“We present DreamTrue, a multi-view, cross-embodiment robot world model for action-faithful and physically plausible video predi…”

They claim — On AgiBot, DreamTrue attains state-of-the-art action following, while reducing the human-assessed interaction defect rate from from 48.12% to 6.25%.

Published 2026-10-07

Fetched 2026-10-09 23:01 UTC · 074b3efb8c207b56

Torn from — the abstract

“Egocentric world models predict first-person observations conditioned on an agent's actions, but most focus on a single agent. …”

Published 2026-10-07

Fetched 2026-10-09 23:01 UTC · 984783c99f6c9c63

Torn from — the abstract

“Recent 3D world models generate photorealistic, explorable scenes that remain frozen in time. OuroWorld is a mask-free framewor…”

They claim — On 39 reconstructed and generated scenes, OuroWorld outperforms all baselines and wins 70.8%-99.0% of user-study comparisons.

Published 2026-10-07

Fetched 2026-10-09 23:01 UTC · 2df83b14d4e56c3d

Torn from — the abstract

“This is a book about large language models. As indicated by the title, it primarily focuses on foundational concepts rather tha…”

Published 2026-10-07

Fetched 2026-10-09 23:01 UTC · 5a28cbe0f0dba767

Torn from — the abstract

“Learning to act in unfamiliar environments requires agents to infer how the world works and revise that understanding as new ev…”

They claim — On ARC-AGI-3, the single-model agent clears every level of all 25 public games, achieves a mean Relative Human Action Efficiency (RHAE) of 100.0, and uses 44% of the human action count.

Published 2026-10-07

Fetched 2026-10-09 23:01 UTC · 2ec664cc3165cff7

Torn from — the abstract

“Dense correspondence matching has historically been bounded by simplifying spatio-temporal priors, such as smooth motion and ri…”

Published 2026-10-07

Fetched 2026-10-09 23:01 UTC · 928c5916c8745d83

Torn from — the abstract

“Reinforcement learning (RL) is the central training paradigm for advancing large foundation models towards self-improvement. Th…”

They claim — We scale RL compute along three dimensions: (1) larger batches and higher throughput, with an asynchronous training that consumes 1,568 samples and 2.7-3.7B tokens per step at context len…

Published 2026-10-07

Fetched 2026-10-09 23:01 UTC · 1b5db0f1008c3cfe

Torn from — the abstract

“Large language model (LLM) routing distributes inference work across different models, advancing the cost-quality Pareto fronti…”

They claim — Across diverse routing algorithms, workloads, and model pairs, TokenRouter achieves 2.01-64.15x higher decoding throughput than existing systems, substantially advancing the serving effic…

Published 2026-10-07

Fetched 2026-10-09 23:01 UTC · e1c4e3f59be4b797

Torn from — the abstract

“Long-horizon agents compact their history to continue within a finite context window, but a textual summary alone may not suppo…”

They claim — On SummHay, REMORY improves source attribution at nearly unchanged insight coverage and approaches the full-context joint score using only 5.2% of the input positions.

Published 2026-10-07

Fetched 2026-10-09 23:01 UTC · d6956d1b7ab9450f

Torn from — the abstract

“Learning from experience is essential for LLM agents to adapt to unfamiliar and dynmaic environments. Evaluating this ability i…”

They claim — (2) Human agent gap: Top-performing human players achieve higher peak scores than the evaluated agents.

Published 2026-10-07

Fetched 2026-10-09 23:01 UTC · 2a9358bd3d19b21d

Torn from — the abstract

“Large language models are informing decisions with ever-higher stakes. As the consequences of their errors grow, a central ques…”

Published 2026-10-05

Fetched 2026-10-09 23:01 UTC · b917d3ea1ea8f81a

Torn from — the abstract

“We study robotic in-context learning (ICL), an emerging paradigm that enables robots to infer and execute tasks from visual dem…”

Published 2026-09-28

Fetched 2026-10-09 23:01 UTC · 8e9c091792f18ffc

Torn from — the abstract

“Representation autoencoders (RAEs) reuse features from a pretrained visual encoder as reconstruction and diffusion latents, int…”

They claim — On ImageNet-256 with DINOv3-L, a single FuseReg decoder reconstructs from full, sparse, and single-layer fusions without retraining, achieving higher PSNR than decoders specialized to fix…

Published 2026-09-24

Fetched 2026-09-28 05:05 UTC · 6861a4d455b6989e

Torn from — the abstract

“World Action Models (WAMs) jointly model visual dynamics and action generation for generalist robot manipulation. A central cha…”

They claim — We introduce InternW0-Δ, a unified WAM pretrained on a heterogeneous corpus that outperforms prior methods across simulation benchmarks and real-robot platforms.

Published 2026-09-24

Fetched 2026-09-28 05:05 UTC · fce8fc3803ad1364

Torn from — the abstract

“Scaling language models to long contexts is limited by the quadratic cost of self-attention. Block sparse attention offers an e…”

Published 2026-09-24

Fetched 2026-09-28 05:05 UTC · e8ca90729f4aeb76

Torn from — the abstract

“The rapid progression of large language models is extending AI from passive content generation into the active workflows of eng…”

Published 2026-09-23

Fetched 2026-09-25 23:01 UTC · 4f969e088a867b41

Torn from — the abstract

“Task and motion planning (TAMP) problems remain difficult even with full observability and object-centric states because discre…”

They claim — Across all program synthesis methods, we evaluate 980 generated programs on 100 held-out instances each, 98,000 evaluation episodes in total.

Published 2026-09-23

Fetched 2026-09-25 23:01 UTC · 8f9867a434213048

Torn from — the abstract