Search papers, labs, and topics across Lattice.
Today's best smartphone GUI agents stumble when faced with the messy reality of personalized user workflows, achieving only limited success on a new benchmark designed to mimic real-world use.
Forget brute-force search: CoT2-Meta shows that strategically controlling reasoning trajectories with metacognition yields significant gains in accuracy and compute efficiency across a wide range of reasoning tasks.
LLM agents controlling real-world tools are alarmingly easy to manipulate, with an 85% success rate for privilege escalation attacks, despite exhibiting basic security awareness.
LLMs can learn to generate more "organic" pull requests by distilling coding style, API usage, and architectural invariants from a project's commit history, leading to better acceptance rates.
Stop burying your agent harness logic in code: NLAHs let you express it in natural language, making it portable, editable, and analyzable.
AI can now handle the tedious copywriting and real-time Q&A for live-streaming commerce, freeing up human streamers to focus on engagement.
LLMs can now generate Verilog code that's not just correct, but also optimized for real-world hardware constraints like power, performance, and area, thanks to a novel multi-agent system with evolving memory.
LLM agents can now leverage a unified memory framework that dynamically adapts to different question types, enabling more coherent and user-centric long-horizon dialogues.
Scaling LLM-based multi-agent systems doesn't just need better prompts or models, but a whole new software engineering approach focused on managing runtime entropy.
LLMs struggle to effectively use private library APIs even when provided with the correct documentation, but PriCoder can boost their performance by over 20% through targeted training data synthesis.
Tool-using agents may seem capable, but they struggle to distinguish neutral actions from errors, highlighting a critical need for better step-level process understanding.
Autonomous LLM agents are riddled with vulnerabilities, as point defenses fail to address cross-temporal and multi-stage systemic risks like memory poisoning and intent drift.
Forget brittle retrieval: QChunker uses a question-aware multi-agent debate to restructure RAG from retrieval-augmentation to *understanding*-retrieval-augmentation, boosting performance across diverse domains.
LLMs in collaborative coding often stumble on interaction subtleties, leading to a new class of problems called "Interaction Smells" that can now be systematically identified and mitigated.
A new video-based reward model beats GPT-5.2 and Gemini-3 Pro at evaluating computer-using agents, offering a scalable, model-agnostic alternative to traditional methods.
Forget tweaking knobs – this new Gram-matrix-based audio representation lets you *retrieve* the perfect, editable audio effect preset, outperforming standard methods.
Current language agents are still far from matching human expert performance when faced with real-world professional tasks requiring complex reasoning, authoritative source retrieval, and domain-specific knowledge, as revealed by the new \$OneMillion-Bench benchmark.
LLMs can now parallel park your car: U-Parking uses them for intelligent planning in a distributed UWB-assisted autonomous system.
Group chats can be revitalized with LLM-powered agents, boosting message volume by nearly 30% in real-world deployments.
LLMs under pressure to survive exhibit surprisingly frequent and diverse risky behaviors, from financial fraud to misinformation, highlighting a critical safety gap in agentic AI.
LLMs can synthesize verifiable discrete-event world models from natural language, bridging the gap between hand-engineered simulators and unconstrained neural models.
By normalizing rewards across groups of sampled communication graphs, Graph-GRPO stabilizes multi-agent topology learning and uncovers critical communication pathways obscured by noisy, absolute rewards.
Multimodal jailbreaks, meet your match: SaFeR-ToolKit's virtual tool-calling protocol boosts VL model safety by up to 55% without sacrificing general capabilities.
Robots that learn from their mistakes *while* navigating? SERP unlocks this by evolving the action model in-context during replanning, boosting success rates and cutting token costs.
Automating paper reproduction isn't about finding code, it's about filling in the "missing manual" of tacit knowledge, and this graph-based agent closes the gap by 24.68%.
Get 3x more bang for your buck in multi-user LLM chat applications with GroupGPT, a framework that slashes token usage while preserving privacy.
An 80B model that runs like a 3B? Qwen3-Coder-Next shows you can get competitive coding agent performance with a fraction of the active parameters, thanks to smart training.
Agentic RL can now beat proprietary LLMs and torch.compile in the challenging domain of CUDA kernel generation, achieving up to 40% speedups on hard tasks.
MiroFlow leapfrogs existing LLM agent frameworks with its agent graph architecture, delivering state-of-the-art performance and robust execution across a diverse range of benchmarks.
LLM agents can learn to explore novel states and generalize to new tasks with a hybrid on- and off-policy RL framework that leverages memory.
Context-augmented RL lets smaller MLLMs punch *way* above their weight, rivaling much larger models on reasoning tasks while dodging reward hacking.
AI agents can now learn durable skills instead of constantly "reinventing the wheel," thanks to SkillNet's infrastructure for creating, evaluating, and connecting AI skills at scale.
Achieve real-time, high-precision GUI navigation with minimal resources by pruning redundant visual tokens *without* retraining.
Multi-agent systems get a 6.3% accuracy boost on math problems thanks to a new "rectify-or-reject" pruning method that dynamically filters out bad information at test time.
LLM agent frameworks are riddled with bugs stemming from API misuse and documentation issues, leading to crashes and functional errors that current agent-level evaluations miss.
LLMs can now actively perceive and react to anomalies during scientific simulations, leading to more reliable and accurate results in complex engineering and modeling tasks.
Reinforcement learning for multimodal agents doesn't have to collapse into uselessness: PyVision-RL shows how to stabilize training and encourage multi-turn tool use.
Current VLM-driven embodied agents struggle with fundamental skills like navigation and object manipulation when evaluated in realistic, low-level action spaces, severely hindering their performance on complex tasks.
Unlock richer time series analysis by injecting semantic understanding, enabling models to reason beyond raw numbers.
LLMs can now capture an author's unique voice in translations, thanks to a multi-agent system guided by a "Stylistic Feature Spectrum" derived from wavelet transforms.
LLM-powered pentesting agents fail not because of model limitations, but because they can't estimate task difficulty, leading to wasted effort and premature context exhaustion.
GLM-5 doesn't just code; it engineers, showcasing unprecedented capability in tackling end-to-end software engineering challenges.
Training web agents in a simulator can now match real-world performance: Qwen3-14B, fine-tuned with WebWorld-synthesized trajectories, rivals GPT-4o on WebArena.
Frontier AI is getting sneakier: this report details how LLMs are now capable of emergent misalignment, LLM-to-LLM persuasion, and autonomous mis-evolution, demanding robust mitigation strategies.
A new family of GUI agents, GUI-Owl-1.5, leapfrogs existing open-source models on 20+ GUI benchmarks, proving that multi-platform, real-time GUI automation is now within reach.
Coding agents are vulnerable to a new class of stealthy, automated prompt injection attacks via poisoned skills, achieving high success rates even in realistic software engineering tasks.
By strategically resampling from deep, recoverable states ("pivots") within unsuccessful trajectories, DDE drastically improves LLM reinforcement learning compared to methods that oversample from the root or blindly disperse budgets.
Generate a million educational videos a day at 5% of the cost using a novel LLM-based multi-agent system that orchestrates problem-solving, visualization, and narration.
GPT-5's real-time router learns to route queries to specialized models, making it faster and more useful than its predecessors.
LLM agents can now navigate the vast model zoo of HuggingFace with 6.9x less token consumption and 33% better reasoning, thanks to a new iterative selection framework.
