Papers

“Scientific discovery begins where known problems end. There, AI systems must engage in exploration: framing hypotheses, designi…”

They claim — We evaluate 10 AI systems and find that the strongest systems can acquire and apply unfamiliar rules, while performance varies substantially across trajectories and continued exploration …

Published 2026-09-23

Fetched 2026-09-25 23:01 UTC · 9ccf356c7f3c30e0

Torn from — the abstract

“Video generation begins in text space by authoring a cinematic screenplay, then materializes into pixels. As contemporary video…”

Published 2026-09-23

Fetched 2026-09-25 23:01 UTC · 01decd428e012e3f

Torn from — the abstract

“Rufus-Air is an open and reproducible post-training recipe on GLM-4.5-Air-Base (106B-A12B), organized as a serial pipeline of e…”

They claim — Rufus-Air improves over the official GLM-4.5-Air post-trained release and is competitive with similarly sized open models.

Published 2026-09-23

Fetched 2026-09-25 23:01 UTC · ecf373efc650dcaf

Torn from — the abstract

“While Large Language Models (LLMs) rely on highly non-linear components, in this work we demonstrate that they exhibit fundamen…”

Published 2026-09-23

Fetched 2026-09-25 23:01 UTC · 859a30c4317d1a6d

Torn from — the abstract

“Recently, Large Language Models (LLMs) have been increasingly able to solve advanced mathematical problems, including many that…”

They claim — Optimizing for our metric creates a model capable of producing more interesting theorems, while also reducing substantial or full overlap with Mathlib from 91.9% to 30.6%, showcasing the …

Published 2026-09-22

Fetched 2026-09-25 23:01 UTC · 6d69919b7b03eed9

Torn from — the abstract

“Object permanence and solidity are hallmarks of human cognitive priors. Recent studies show that video generation models, a par…”

Published 2026-09-22

Fetched 2026-09-25 23:01 UTC · 9ddf4b0029a949c3

Torn from — the abstract

“Humans can effortlessly localize the direction of a sound source and integrate it with visual cues for reasoning, yet this rema…”

Published 2026-09-19

Fetched 2026-09-25 23:01 UTC · ebf4594fa0d2f86a

Torn from — the abstract

“Foundation models, alongside advances in learned game-world models, are reshaping AI across the game lifecycle. Beyond playing …”

Published 2026-09-14

Fetched 2026-09-16 23:01 UTC · bcd23ad210dcebdc

Torn from — the abstract

“PhysBrain 1.5 unifies physical environment understanding, action generation, and future state prediction via joint autoregressi…”

They claim — Across 28 embodied understanding benchmarks, our 8B model achieves an average score of 72.5, setting a new open-source state of the art and performing on par with leading proprietary mode…

Published 2026-09-13

Fetched 2026-09-15 23:01 UTC · 70061e1a1aeb68ea

Torn from — the abstract

“Dream-RSI enables scalable recursive self-improvement by using historical discovery replay to evaluate exploration policies off…”

Published 2026-09-13

Fetched 2026-09-15 23:01 UTC · 263aacf523ded915

Torn from — the abstract

“Atria Dawn Preview is a foundation agentic language model trained through verified tool interactions that achieves strong bench…”

They claim — Across 16 benchmarks spanning real-world research, engineering, and digital work, Atria Dawn Preview is competitive with frontier agents and achieves the highest reported score on five of…

Published 2026-09-13

Fetched 2026-09-15 23:01 UTC · 21c391716120f827

Torn from — the abstract

“Realtime spoken interaction demands deep reasoning, prompt responses, and fluid turn-taking. We present StepAudio 3 Realtime, a…”

They claim — StepAudio 3 Realtime achieves top-tier performance across key dimensions: an exceptional 90.6 on the MMSU benchmark, 98.9 Overall on the Artificial Analysis Full-Duplex Bench, and a 56.0%…

Published 2026-09-11

Fetched 2026-09-16 23:01 UTC · c13c53fb82422478

Torn from — the abstract

“LIT improves robot action generalization by first training pose-conditioned action priors without images, then constraining vis…”

They claim — Across four vision-language-action and world-action architectures (Pi0.5, MolmoAct2, FAST-WAM, and ImageWAM), LIT improves overall LIBERO-Plus success by 3.87-10.70 percentage points whil…

Published 2026-09-10

Fetched 2026-09-14 23:01 UTC · 3e83d75847ca11ec

Torn from — the abstract

“ZGCM-1 is a 7B open foundation model that combines internal reasoning with external tool use, trained via efficient architectur…”

They claim — On several challenging mathematical reasoning and agentic search suites, it remains competitive with frontier models orders of magnitude larger, such as Qwen3-235B-A22B and GLM-5.1.

Published 2026-09-10

Fetched 2026-09-15 23:01 UTC · fcb1e94d9a32e870

Torn from — the abstract

“Benchmark Radar is a searchable living database and discovery engine for AI evaluation benchmarks that aggregates sources, scor…”

Published 2026-09-09

Fetched 2026-09-14 23:01 UTC · 1cc8c83680c0c992

Torn from — the abstract

“Vidu S2 introduces real-time interactive avatar and video editing models that support high-resolution spatial video generation …”

They claim — Experiments show that Vidu S2 outperforms all baselines.

Published 2026-09-09

Fetched 2026-09-15 23:01 UTC · 5e45abe94152d970

Torn from — the abstract

“A data-centric framework with specialized systems for reasoning analysis, cost reduction, and execution verification enables sm…”

They claim — The three checkpoints improve over their starting models by an average of 23.76% on the full CyberGym suite and 10.49% across the pooled CTF suites.

Published 2026-09-07

Fetched 2026-09-14 23:01 UTC · 4d29dafb2e864139

Torn from — the abstract

“Composing complementary continual learning mechanisms substantially improves long-horizon memorization in language models by co…”

They claim — Our best method combines all three anchors with merged LoRA, ranks among the top 3 methods in all datasets, and raises average final retention from 1.2% under naive sequential fine-tuning…

Published 2026-09-06

Fetched 2026-09-16 23:01 UTC · 299fad04b14317fa

Torn from — the abstract

“DataFlex-RL evaluates reinforcement learning data policies and finds that uniform sampling matches or exceeds adaptive rollout …”

They claim — Uniform GRPO improves the domain-balanced average accuracy by 7.76 percentage points over the untrained checkpoint.

Published 2026-09-04

Fetched 2026-09-14 23:01 UTC · 8b008970febe99b9

Torn from — the abstract

“Layer dropout (a.k.a. stochastic depth) has been shown to enable faster training, higher accuracy, and robustness to zero-shot …”

They claim — For a given number of training steps, LLMs can achieve lower or similar validation loss while saving upto 25% of training FLOPs.

Published 2026-09-04

Fetched 2026-09-07 01:32 UTC · 60c554528e9aa388

Torn from — the abstract

“Large language models (LLMs) are increasingly used to formulate optimization models from natural-language problem descriptions,…”

Published 2026-09-04

Fetched 2026-09-07 01:32 UTC · 1ba5a6f5e64b9cda

Torn from — the abstract

“Adapting a single-object 3D generative prior to multi-view observations enables scalable compositional mesh reconstruction of d…”

Published 2026-09-03

Fetched 2026-09-07 23:01 UTC · acfc206edb29736d

Torn from — the abstract

“UniMate is a unified diffusion transformer that generates articulated motion for arbitrary skeletons from text and rigged 3D as…”

They claim — UniMate introduces a topology-aware diffusion transformer, which integrates skeletal topology into attention via three mechanisms: (1) a graph-aware attention bias from pairwise joint rel…

Published 2026-09-03

Fetched 2026-09-07 23:01 UTC · f7aead1ac8cf4d5a

Torn from — the abstract

“Distinct reasoning operations in language models are geometrically separable in hidden representations, with structure emerging…”

Published 2026-09-03

Fetched 2026-09-07 23:01 UTC · 8fd8b68e8d2fd445

Torn from — the abstract

“Two large-scale search agents are trained via a multi-stage pipeline combining supervised fine-tuning and reinforcement learnin…”

Published 2026-09-02

Fetched 2026-09-07 23:01 UTC · bf933da181c11d7a

Torn from — the abstract

“Audio-video diffusion models exhibit bidirectional semantic leakage through cross-modal attention pathways, which can be diagno…”

Published 2026-09-02

Fetched 2026-09-07 23:01 UTC · 496e910644d364a4

Torn from — the abstract

“ENEAS unifies text-prompted instance tracking and open-concept semantic discovery via temporal memory extension and a verificat…”

Published 2026-09-02

Fetched 2026-09-08 16:21 UTC · a5384a599ed5599b

Torn from — the abstract

“FlowBalance improves reasoning models via verifier-calibrated self-guidance using trajectory-level score reweighting and profil…”

They claim — On mathematical reasoning, FlowBalance improves average performance over FlowRL on both Qwen3-4B and Qwen3-8B, while also improving training speed and stability, avoiding direct OPSD's re…

Published 2026-09-02

Fetched 2026-09-08 16:21 UTC · 437efcb18b365a17

Torn from — the abstract

“Enoki is an open information extraction framework that unifies claim-level verification and span-level hallucination localizati…”

Published 2026-08-31

Fetched 2026-09-07 23:01 UTC · ab91839c3bae8764

Torn from — the abstract

“Command-line coding agents (e.g., Claude Code, Gemini CLI) can already read and write files and sustain long sessions, yet end-…”

Published 2026-08-30

Fetched 2026-09-07 23:01 UTC · 17101143ec4ae6bc

Torn from — the abstract