CHASM is a new benchmark dataset showing that existing multimodal large language models fail to reliably detect covert advertisements on Chinese social media even after fine-tuning.
super hub Mixed citations
ChatGLM: A Family of Large Language Models from GLM-130B to GLM-4 All Tools
Mixed citation behavior. Most common role is background (58%).
116
Pith papers citing it
Background
58% of classified citations
abstract
We introduce ChatGLM, an evolving family of large language models that we have been developing over time. This report primarily focuses on the GLM-4 language series, which includes GLM-4, GLM-4-Air, and GLM-4-9B. They represent our most capable models that are trained with all the insights and lessons gained from the preceding three generations of ChatGLM. To date, the GLM-4 models are pre-trained on ten trillions of tokens mostly in Chinese and English, along with a small set of corpus from 24 languages, and aligned primarily for Chinese and English usage. The high-quality alignment is achieved via a multi-stage post-training process, which involves supervised fine-tuning and learning from human feedback. Evaluations show that GLM-4 1) closely rivals or outperforms GPT-4 in terms of general metrics such as MMLU, GSM8K, MATH, BBH, GPQA, and HumanEval, 2) gets close to GPT-4-Turbo in instruction following as measured by IFEval, 3) matches GPT-4 Turbo (128K) and Claude 3 for long context tasks, and 4) outperforms GPT-4 in Chinese alignments as measured by AlignBench. The GLM-4 All Tools model is further aligned to understand user intent and autonomously decide when and which tool(s) touse -- including web browser, Python interpreter, text-to-image model, and user-defined functions -- to effectively complete complex tasks. In practical applications, it matches and even surpasses GPT-4 All Tools in tasks like accessing online information via web browsing and solving math problems using Python interpreter. Over the course, we have open-sourced a series of models, including ChatGLM-6B (three generations), GLM-4-9B (128K, 1M), GLM-4V-9B, WebGLM, and CodeGeeX, attracting over 10 million downloads on Hugging face in the year 2023 alone. The open models can be accessed through https://github.com/THUDM and https://huggingface.co/THUDM.
hub tools
citation-role summary
background 14
baseline 9
method 2
other 1
citation-polarity summary
claims ledger
- abstract We introduce ChatGLM, an evolving family of large language models that we have been developing over time. This report primarily focuses on the GLM-4 language series, which includes GLM-4, GLM-4-Air, and GLM-4-9B. They represent our most capable models that are trained with all the insights and lessons gained from the preceding three generations of ChatGLM. To date, the GLM-4 models are pre-trained on ten trillions of tokens mostly in Chinese and English, along with a small set of corpus from 24 languages, and aligned primarily for Chinese and English usage. The high-quality alignment is achiev
authors
co-cited works
representative citing papers
DDIPE poisons LLM agent skills by embedding malicious logic in documentation examples, achieving 11.6-33.5% bypass rates across frameworks while explicit attacks are blocked, with 2.5% evading detection.
VLRS-Bench is the first benchmark dedicated to complex vision-language reasoning in remote sensing, with 2000 QA pairs across 14 tasks in cognition, decision, and prediction dimensions.
ErrorRadar is a new benchmark of 2,500 multimodal K-12 math problems for MLLM error step identification and categorization, where GPT-4o trails human experts by ~10%.
AirGroundBench is a new diagnostic benchmark exposing that MLLMs handle basic spatial perception but struggle with cross-view alignment, transformation reasoning, and embodied navigation under heterogeneous air-ground views.
CrypFormBench is a new benchmark jointly covering symbolic and computational security to evaluate LLMs on five formal analysis capabilities, with results showing top model Claude-3.5 scores 48.7/100 and most models struggling on generation, transformation, and correction.
MetaSyn benchmark shows LLM agents recover at most 52.7% of relevant studies in meta-analysis pipelines due to failures in PI/ECO-based screening despite strong retrieval.
RWGBench is a citation-centric benchmark for related work generation built from 40k CS papers and a 100-paper test set, with multi-dimensional metrics that better match human expert judgment than standard similarity scores.
Moral Trolley Arena shows frontier LLMs produce composite moral preferences that are compressed rather than additive functions of calibrated component act strengths across Moral Foundations Theory.
Introduces the Grounded Personality Reasoning task and MM-OCEAN dataset to show that MLLMs frequently produce correct Big Five personality ratings without grounding them in observable video evidence.
Text2CAD-Bench supplies 600 dual-prompt examples across four geometric and domain levels to test LLMs on text-to-parametric CAD, finding solid basic performance but sharp drops on complex topology and advanced features.
GeoVista introduces a planning-driven active perception framework with global exploration plans, branch-wise local inspection, and explicit evidence tracking to achieve state-of-the-art results on ultra-high-resolution remote sensing benchmarks.
Language models show a scale-dependent switch from anticorrelated to correlated reasoning-truthfulness coupling at a family-specific critical parameter count, with architecture and data choices shifting the transition point.
PRISM is a tiered benchmark with 300 human-verified tasks across five photorealistic apartments that diagnoses embodied agent failures in basic ability, reasoning ability, and long-horizon ability using an agent-agnostic API.
K12-KGraph is a textbook-derived knowledge graph that powers a new benchmark revealing LLMs' poor curriculum cognition and a small training corpus that outperforms general instruction data on educational tasks.
LMMs perceive videos but underexploit visual content for causal reasoning due to textual shortcuts; ProCauEval diagnoses this and ADPO training reduces reliance on priors.
VITA-QinYu is the first expressive end-to-end spoken language model supporting role-playing and singing alongside conversation, trained on 15.8K hours of data and outperforming prior models on expressiveness and conversational benchmarks.
Tutti is a GPU-direct SSD-backed KV cache that removes CPU bottlenecks via object abstraction, GPU io_uring, and slack scheduling, delivering near-DRAM performance at 2x higher request rate and 27% lower cost than prior GDS-based systems.
OralMLLM-Bench reveals performance gaps between multimodal large language models and clinicians on cognitive tasks for dental radiographic analysis across periapical, panoramic, and cephalometric images.
FinSafetyBench shows that LLMs remain vulnerable to adversarial prompts that bypass financial compliance safeguards, with notably higher failure rates in Chinese-language scenarios.
AuDisAgent reformulates multimodal controversy detection as a dynamic audience dissemination process using screening, panel discussion, and arbitration agents, plus comment bootstrapping, and reports outperforming prior static methods on a public dataset.
ShredBench shows state-of-the-art MLLMs perform well on intact documents but suffer sharp drops in restoration accuracy as fragmentation increases to 8-16 pieces, indicating insufficient cross-modal semantic reasoning for VRDU.
EmoTrans is a new video benchmark with four progressive tasks that measures how well current multimodal LLMs handle dynamic emotion transitions rather than static recognition.
Introduces culture-aware humorous captioning task and staged alignment framework that improves contextual fit and balances image relevance with humor in multimodal LLMs.
citing papers explorer
-
On the Role of Language Representations in Auto-Bidding: Findings and Implications
SemBid injects LLM-encoded Task, History, and Strategy semantics as tokens into offline bidding trajectories and uses self-attention to outperform numerical-only baselines in performance, constraint satisfaction, and robustness.
-
CAR: Query-Guided Confidence-Aware Reranking for Retrieval-Augmented Generation
CAR reranks documents in RAG by promoting those that increase generator confidence (via answer consistency sampling) and demoting those that decrease it, yielding NDCG@5 gains on BEIR datasets that correlate with F1 improvements.
-
Theory-Grounded Evaluation Exposes the Authorship Gap in LLM Personalization
Theory-grounded authorship metrics show four LLM personalization methods score below calibrated baselines (0.484-0.508 vs. 0.626 floor), exposing a gap hidden by uncalibrated evaluations.
-
Salca: A Sparsity-Aware Hardware Accelerator for Efficient Long-Context Attention Decoding
Salca is a new ASIC accelerator that achieves 3.82× speedup and 74.19× energy efficiency over A100 for long-context attention via dual-compression dynamic sparse attention and pipelined hardware.
-
CAP: Controllable Alignment Prompting for Unlearning in LLMs
CAP is a reinforcement-learning-driven prompt optimization framework that suppresses target knowledge in LLMs while preserving general capabilities, enabling reversible unlearning without any parameter updates.
-
LLM Safety From Within: Detecting Harmful Content with Internal Representations
SIREN identifies safety neurons via linear probing on internal LLM layers and combines them with adaptive weighting to detect harm, outperforming prior guard models with 250x fewer parameters.
-
Multi-LLM Token Filtering and Routing for Sequential Recommendation
MLTFR combines user-guided token filtering with a multi-LLM mixture-of-experts and Fisher-weighted consensus expert to deliver stable gains in corpus-free sequential recommendation.
-
MemExplorer: Navigating the Heterogeneous Memory Design Space for Agentic Inference NPUs
MemExplorer optimizes heterogeneous memory systems for agentic LLM inference on NPUs and reports up to 2.3x higher energy efficiency than baselines under fixed power budgets.
-
The Salami Slicing Threat: Exploiting Cumulative Risks in LLM Systems
Salami Attack chains low-risk inputs to cumulatively trigger high-risk LLM behaviors, achieving over 90% success on GPT-4o and Gemini while resisting some defenses.
-
Adapting 2D Multi-Modal Large Language Model for 3D CT Image Analysis
Transferring a 2D MLLM to 3D CT inputs via parameter reuse, a Text-Guided Hierarchical MoE framework, and two-stage training yields better performance than prior 3D medical MLLMs on medical report generation and visual question answering.
-
Aligned Agents, Biased Swarm: Measuring Bias Amplification in Multi-Agent Systems
Multi-agent systems amplify minor stochastic biases into systemic polarization via echo-chamber effects in structured workflows, even with neutral agents.
-
The Art of (Mis)alignment: How Fine-Tuning Methods Effectively Misalign and Realign LLMs in Post-Training
ORPO is most effective at misaligning LLMs while DPO excels at realigning them, though it reduces utility, revealing an asymmetry between attack and defense methods.
-
In-Place Test-Time Training
In-Place TTT adapts LLM MLP projection matrices at test time with a next-token-aligned objective and chunk-wise updates, enabling better long-context performance as a drop-in enhancement.
-
Cooking Up Risks: Benchmarking and Reducing Food Safety Risks in Large Language Models
A new benchmark exposes food-safety gaps in current LLMs and guardrails, and a fine-tuned 4B model is offered as a domain-specific fix.
-
Ensemble-Based Uncertainty Estimation for Code Correctness Estimation
Ensemble Semantic Entropy improves correlation with code correctness over single-model methods and powers a cascading scaling system that cuts FLOPs by 64.9% while preserving performance on LiveCodeBench.
-
Attention Sinks Induce Gradient Sinks: Massive Activations as Gradient Regulators in Transformers
Attention sinks induce gradient sinks under causal masking, with massive activations serving as adaptive RMSNorm regulators that attenuate localized gradient pressure in Transformer training.
-
Performance Isolation and Semantic Determinism in Efficient GPU Spatial Sharing
CoGPU resolves the tradeoff in GPU sharing by introducing GPU coroutines for semantic-preserving resource migration, delivering up to 79.2% higher training throughput and zero token mismatch in inference.
-
RPC-Bench: A Fine-grained Benchmark for Research Paper Comprehension
RPC-Bench supplies 15K verified QA pairs and a research-flow taxonomy that shows top foundation models still achieve only 68.2 percent correctness-completeness on academic paper comprehension.
-
Listen to Rhythm, Choose Movements: Autoregressive Multimodal Dance Generation via Diffusion and Mamba with Decoupled Dance Dataset
LRCM is a new multimodal diffusion model with audio and text Conformers plus Motion Temporal Mamba for generating long, coherent dance sequences from rhythm and descriptions using a decoupled dataset.
-
CascadeInfer: Length-Aware Scheduling of LLM Serving with Low Latency and Load Balancing
CascadeInfer partitions LLM instances into length-specialized groups, uses dynamic programming for stage partitioning, and applies runtime refinement plus decentralized load balancing to cut latency and raise throughput.
-
MEASER: Malware embedding attacks on open-source LLMs
MEASER embeds malware into open-source LLMs via parameter targeting and MAR-QIM modulation, achieving 0 BER and high stealth even after quantization and PEFT.
-
SCOPE-RL: Stable and Quantitative Control of Policy Entropy in RL Post-Training
SCOPE-RL adds a regularization term built from high-temperature positive samples to quantitatively control entropy dynamics and maintain exploration in RL post-training of reasoning LLMs.
-
Mitigating Visual Context Degradation in Large Multimodal Models: A Training-Free Decoupled Agentic Framework
DRP decouples reasoning from perception in LMMs by using an LLM reasoner to query an LMM observer for visual details as needed, reducing visual grounding loss.
-
GLM-4.5V and GLM-4.1V-Thinking: Towards Versatile Multimodal Reasoning with Scalable Reinforcement Learning
GLM-4.5V reaches state-of-the-art results on 42 multimodal benchmarks among open-source models of similar size by applying reinforcement learning with curriculum sampling to a strong vision foundation model.
-
LongWriter-Zero: Mastering Ultra-Long Text Generation via Reinforcement Learning
LongWriter-Zero applies RL from a base model with specialized rewards for length, quality, and structure to outperform SFT baselines and larger models on long-writing benchmarks.
-
Real-World Doctor Agent with Proactive Consultation through Multi-Agent Reinforcement Learning
DoctorAgent-RL trains a Qwen2.5-7B doctor agent via multi-agent RL on the new MTMedDialog dataset to conduct dynamic, question-driven consultations, reaching 70% exact diagnostic match in real-patient trials.
-
Qwen2.5-1M Technical Report
Qwen2.5-1M models reach 1M token context with improved long-context performance, no short-context loss, and 3-7x prefill speedup via open inference optimizations.
-
LongBench v2: Towards Deeper Understanding and Reasoning on Realistic Long-context Multitasks
LongBench v2 benchmark shows current LLMs underperform humans on deep long-context reasoning tasks, but extended inference-time reasoning enables surpassing the human baseline.
-
GLM-4-Voice: Towards Intelligent and Human-Like End-to-End Spoken Chatbot
GLM-4-Voice builds an end-to-end spoken chatbot by deriving a 175bps single-codebook tokenizer from ASR, synthesizing interleaved speech-text data, and continuing pre-training of GLM-4-9B on up to 1 trillion tokens before fine-tuning on conversational speech.
-
Predict, Reuse, and Repair: Accelerating Dynamic Sparse Attention for Long-Context LLM Decoding
PRR accelerates dynamic sparse attention decoding in long-context LLMs via EMA-based prediction, speculative attention, and FlashAttention repair, achieving up to 40% latency reduction.
-
Representation Collapse in Sequential Post-Training of Large Language Models
Sequential post-training of LLMs induces representation collapse that correlates with reduced plasticity, weaker generalization, and poorer calibration, with lightweight interventions tested to mitigate it.
-
Mobile-Aptus: Confidence-Driven Proactive and Robust Interaction in MLLM-based Mobile-Using Agents
Mobile-Aptus uses supervised fine-tuning followed by semantic similarity retrieval and direct preference optimization to calibrate confidence scores in mobile agents, yielding over 17% average task success improvement on four benchmarks.
-
Active Evidence-Seeking and Diagnostic Reasoning in Large Language Models for Clinical Decision Support
Multi-turn evidence seeking reduces LLM diagnostic accuracy by 12.75% and supporting-evidence quality by 24.36% versus full-context evaluation in a new OSCE-inspired benchmark across 468 cases and 15 models.
-
VIPER-MCP: Detecting and Exploiting Taint-Style Vulnerabilities in Model Context Protocol Servers
VIPER-MCP detects and exploits taint-style vulnerabilities in Model Context Protocol servers via anchor-query static analysis and feedback-driven prompt evolution, uncovering 106 zero-day vulnerabilities across 39,884 repositories with 67 CVEs assigned.
-
ECG-WM: A Physiology-Informed ECG World Model for Clinical Intervention Simulation
ECG-WM combines ODE physiological priors with latent diffusion models to generate intervention-conditioned ECG trajectories and uses diffusion stochasticity for uncertainty-aware clinical risk assessment.
-
New Wide-Net-Casting Jailbreak Attacks Risk Large Models
The paper demonstrates that a tailored jailbreak method for querying groups of large models can achieve up to 100% success rate in some experiments on unprotected models, revealing overlooked multi-model safety risks.
-
Sakura at BEA 2026 Shared Task 1: What Makes Vocabulary Difficult?
Fine-tuned LLM and explainable models predict vocabulary difficulty with correlations r > 0.91 and r > 0.77, showing spelling difficulty and test item construction as key influences in addition to word production difficulty.
-
When Emotion Becomes Trigger: Emotion-style dynamic Backdoor Attack Parasitising Large Language Models
Paraesthesia is an emotion-style dynamic backdoor attack achieving ~99% success rate on instruction and classification tasks across four LLMs while preserving clean performance.
-
Minimizing Modality Gap from the Input Side: Your Speech LLM Can Be a Prosody-Aware Text LLM
TextPro-SLM reduces the speech-text modality gap by feeding an LLM backbone with synchronized text tokens and prosody embeddings from WhisperPro, achieving lowest gap scores at 3B/7B scales with roughly 1,000 hours of audio.
-
From Coarse to Fine: Self-Adaptive Hierarchical Planning for LLM Agents
AdaPlan-H enables LLM agents to generate self-adaptive hierarchical plans that adjust detail level to task difficulty, improving success rates in multi-step tasks.
-
ReCAPA: Hierarchical Predictive Correction to Mitigate Cascading Failures
ReCAPA adds predictive correction and multi-level semantic alignment to VLA models, plus two new metrics for tracking error spread and recovery, yielding competitive benchmark results over LLM baselines.
-
SparseBalance: Load-Balanced Long Context Training with Dynamic Sparse Attention
SparseBalance dynamically adjusts sparsity and batches workloads to load-balance sparse attention training, delivering up to 1.33x speedup and 0.46% better long-context performance on LongBench.
-
Disposition Distillation at Small Scale: A Three-Arc Negative Result
Multiple standard techniques for instilling dispositions in small LMs consistently failed across five models, with initial apparent gains revealed as artifacts and cross-validation collapsing to chance.
-
MAFIG: Multi-agent Driven Formal Instruction Generation Framework
MAFIG uses a Perception Agent and Emergency Decision Agent plus span-focused local distillation to let lightweight models rapidly generate formal instructions that fix local scheduling failures, achieving over 94% success with sub-second latency on port, warehousing, and deck datasets.
-
A Scalable Multi-LLM Collaboration System with Retrieval-based Selection and Exploration-Exploitation-Driven Enhancement
SMCS coordinates 15 open-source LLMs via retrieval-based prior selection and exploration-exploitation posterior enhancement, outperforming GPT-4.1 by 5.36% and GPT-o3-mini by 5.28% on eight benchmarks.
-
LENS: Multi-level Evaluation of Multimodal Reasoning with Large Language Models
LENS is a new multi-level benchmark dataset for evaluating MLLMs on perception-to-reasoning tasks using the same images across all levels with recent social media content.
-
Advancing AI Research Assistants with Expert-Involved Learning
ARIEL evaluates LLMs and LMMs on full-length biomedical summarization and figure interpretation with blinded expert review, identifies limitations, and demonstrates gains from prompt engineering, fine-tuning, and an integrated agent for hypothesis generation.
-
From Human Memory to AI Memory: A Survey on Memory Mechanisms in the Era of LLMs
The paper surveys human memory categories, maps them to LLM memory, and proposes a new three-dimension (object, form, time) categorization into eight quadrants to organize existing work and highlight open problems.
-
Q-Agent: Quality-Driven Chain-of-Thought Image Restoration Agent through Robust Multimodal Large Language Model
Q-Agent uses CoT decomposition on a fine-tuned MLLM for multi-degradation perception plus IQA-driven greedy selection of restoration algorithms to claim better performance than All-in-One IR models.
-
Towards Reasoning Era: A Survey of Long Chain-of-Thought for Reasoning Large Language Models
The paper unifies perspectives on Long CoT in reasoning LLMs by introducing a taxonomy, detailing characteristics of deep reasoning and reflection, and discussing emergence phenomena and future directions.