AI Dataset2026
Creator-centric benchmark for evaluating text-to-image models with 1,000 bilingual prompts and a 3-level, 56-facet taxonomy. Includes a trained Q-Judger judge model and leaderboard-ready evaluation scripts to surface gaps in real-world fidelity and creative generation.
AI Dataset2026
Provides a 289-case (1,058-turn) multi-turn benchmark that evaluates interactive video world models across 22 metrics and five dimensions (quality, setting, interaction, consistency, physics). Includes first-/third-person and navigation splits plus a 20-model leaderboard for head-to-head comparisons.
Combines internalizing general skills with task-specific skill utilization via a difficulty-aware router to improve in-distribution and out-of-distribution performance for agentic RL. Uses privileged distillation for hard tasks and diagnostic probing for easy tasks; evaluated on ALFWorld and WebShop.
AI Agent Papers2026
Proposes TASTE, an automatic pipeline that synthesizes challenging agent benchmark tasks by sampling and evolving valid tool-sequence patterns; uses an adaptive contrastive n-gram model and LLM validity judgments to produce τ^c-Bench with broader tool-use coverage and higher difficulty.
Analyzes spatial representations in vision–language models and reveals a consistent vertical-position ↔ distance entanglement; introduces SpatialTunnel, a synthetic benchmark that exposes this perspective-driven shortcut, and provides code and a project page.
AI Video Papers2026
Performs training-free early-stage visual token compression inside the vision encoder to cut time-to-first-token (TTFT) and FLOPs for Video-LLMs. Introduces a decoupled spatial token selection strategy and reports up to 2.65× TTFT reduction and 61% FLOPs savings on LLaVA-OneVision-7B (NVIDIA A100) while preserving full-token accuracy — aimed at latency-sensitive video understanding.
AI Dataset2026
A collection of 14,056 self-contained research-level mathematical problems extracted from papers and open-problem lists, each rewritten with taxonomy labels and open-status metadata for training or evaluating models on research-grade math reasoning.
Learns a text-conditioned flow (a conditional velocity field) in LLM residual activations to steer frozen models at inference by partially transporting and regenerating activations under target textual conditions — enabling unified control over persona, style, truthfulness, compositional constraints, and activation-space classification.
AI Video Papers2026
Enables real-time streaming video-to-video editing (1280×704 @24 FPS) on a single RTX 5090 GPU. Uses a Hybrid Diffusion Transformer for balanced local/global modeling, Cycle‑Reverse Regularization for temporal consistency, and system-level mixed-precision and fused kernels to maximize throughput.
Introduces Draft-OPD, an on-policy distillation method for training lightweight draft models used in speculative decoding — it focuses learning on draft-induced errors via target-assisted rollouts and replay, improving acceptance length and enabling >5× lossless LLM inference acceleration.
Reallocates injected noise energy across frequency bands to match a diffusion model's spectral bias, improving sampling fidelity without retraining. Uses a timestep- and frequency-dependent colored-noise schedule as a plug-and-play inference-time SDE solver; shows sizable FID drops on ImageNet-256.
AI Agent Papers2026
Analyzes when masking stale observations improves long-horizon search agents and why, identifying an asymmetric inverted-U relationship between masking benefit, retriever quality, and model capacity; explains a token-for-turn trade-off and releases evaluation scaffolds and trajectories.