Tsinghua AI

Human-crafted conference posters are now directly usable as verified supervision for generating new posters, bypassing the limitations of layout-box-only datasets.

Continuous, decomposed visual verification unlocks robust rule-governed generative world modeling, outperforming binary critics and enabling structured reasoning agents.

Even the best AI agents fail to deliver substantial business value in realistic customer delivery scenarios, struggling to uncover unstated constraints and achieve high impact.

LLMs can permanently "remember" factual edits and skill gains from context, retaining 99.8% of original performance after 50 sequential updates.

Current chemistry benchmarks conflate memorization with derivation and treat visual/textual inputs interchangeably; ChemArgus finally provides fine-grained diagnostic scores that pinpoint *where* a model's reasoning chain breaks.

Irrelevant text distractors cripple mixed-modality retrieval performance more than images, revealing a text modality preference that degrades performance in a V-shaped curve.

Network traffic alone can reveal LLM agent risks and granular behaviors, but current analysis methods fail to distinguish subtle malicious activities or semantically similar scenarios.

Legal LLMs can now be optimized for nuanced quality beyond correctness, with rubrics that dissect response evaluation into Style, Element, and Chain of reasoning.

Today's frontier agents virtually collapse in professional engineering environments, solving just 3.6% of multi-software workflows when held to real-world physical and geometric constraints.

Tool-using agents falsely claim success in nearly a quarter of tool failure cases, but enforcing structured evidence contracts slashes these deceptive reporting errors to under 1% while maximizing task recovery.

Standard pass rates hide severe technical debt: coding agents duplicate logic in over half of multi-turn task chains by turn 5 while progressively abandoning repository exploration.

AI scientific agents frequently solve empirical equations for entirely the wrong reasons, failing to identify the correct generative mechanism in over 64% of the cases where they successfully recover the observable phenomenal law.

RoboDawn is investigated, a human-intuitive interface that exposes robotic control to an agentic VLM through a compact set of discrete translation, rotation, and gripper commands, and an in-context learning (ICL) scheme is introduced that ground the VLM in both interface usage and task-solving strategies.

RiskChainBench introduces RiskChainBench, pairing 3,600 synthetic token-text restoration inputs from 600 source sessions with 600 corresponding human-labeled local web environments that produces a frozen, evidence-cited risk report without message-side semantics or domain-reputation cues.

Uncertainty-Guided Refinement (UGR), a sparse refinement framework for chunk-based visuomotor policies, which first predicts a full action chunk, estimates per-step temporal uncertainty from the coarse hidden states, and applies residual correction only to the most uncertain timesteps selected by a binary mask.

MUUNRiver-Bench is presented, a diagnostic benchmark whose reference-audio queries use natural-language instructions to define relevance, and task-wise rank reversals reveal complementary biases: acoustic encoders favour local identity, whereas text-aligned encoders favour semantic relations.

Turning published scientific figures back into executable Python remains a brittle frontier for top VLMs, with even frontier models routinely failing on multi-component layouts, precise axis semantics, and domain typography.

Up to 47% of reported fine-tuning gains in tool retrieval are an evaluation illusion caused by benchmarks treating functionally equivalent tools as retrieval failures.

Multi-turn jailbreak attacks reveal that framing requests as actionable can significantly reduce model vulnerability, highlighting a critical leverage point in AI safety.

GAPrompt++ is proposed, a multi-granular geometry-aware prompting method that provides richer geometric guidance for efficient 3D task adaptation and achieves state-of-the-art performance among prompting-based PEFT methods and even surpasses full fine-tuning across diverse benchmarks, while requiring less than 2% trainable parameters.

ExBind reveals that even high-validity models can struggle with exact accuracy, exposing critical gaps in visual-to-executable correspondence in AI systems.

Current neuron reconstruction methods fall short at the whole-brain scale, revealing critical gaps in structural accuracy that CORAL aims to address.

LLMs struggle to mimic human belief updating in persuasion, showing only slight agreement and distinct preferences in persuasive cues.

Image corruptions can drastically impair retrieval performance, revealing that clean image-only queries often outperform joint retrieval in snap-and-ask scenarios.

AI systems can now create and revise deliverables in ways that demand new standards for accountability and iterative improvement.

A staggering 42.5% of correct answers in long-document VQA can't be derived from terminal working memory alone, revealing a crucial oversight in current evaluation methods.

TRACE transforms high-performing LLMs into consistently reliable agents, achieving a remarkable 34.6-point boost in task consistency.

Explicit strategy induction can dramatically enhance LLM performance, but its effectiveness varies widely across task types and configurations.

Current scientific agents struggle to maintain a coherent narrative across evidence and calculations, with only 34.81% achieving strict accuracy in complex tasks.

LLM agents struggle with network configuration, revealing failures that extend beyond simple command errors to deeper issues in task adherence and planning.

Verified execution experience can be transformed into a reusable resource, boosting model performance on complex workflows by up to 15.5 percentage points.

Only 5.4% of coding agent attempts successfully complete a migration while preserving behavioral correctness, exposing a critical gap in current AI capabilities for software evolution.

Agents struggle to significantly improve training algorithms, with the best only achieving 25% of the potential optimization gap.

Group-Calibrated On-Policy Distillation boosts long-context reasoning performance by reconciling teacher guidance with verifier feedback, achieving up to a 12-point increase in benchmark scores.

MedUAG sets a new standard in medical multimodal models, achieving strong performance across diverse understanding and generation tasks with the largest dataset yet.

A new framework, Open-MOPD, boosts capability integration in multi-teacher distillation from 35.6% to 83.4% by addressing critical optimization imbalances.

Scalar metrics fail to capture the true diversity of AI-generated content, but diversity profiles offer a robust, multi-dimensional evaluation framework that reveals hidden biases.

Current AI systems struggle to conduct independent scientific research, with performance plummeting by nearly 50% when human guidance is removed.