Every paper checked — nothing hidden

All papers

Every paper checked for brief #4 of 12–19 Oct 2026 — nothing is hidden. Brief papers first.

All17,313 In brief24 More (on-topic)3,333 Not relevant13,754 Not assessed87 Older352 Could not confirm93 Blogs with no paper266 All blogs284

Paper	Topic	Relevance
What Should SFT Learn? Conditional Information Gain for Generalizable Post-Training Introduces an information-theoretic framework to reweight supervised fine-tuning data based on conditional information gain. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Proposes a training method rather than an evaluation benchmark or judge metric.
What Should SFT Learn? Conditional Information Gain for Generalizable Post-Training Introduces an information-theoretic framework to reweight supervised fine-tuning data based on conditional information gain. Preprint	LLMs for code via arxiv · 12 Oct, 16:28	Not relevant General LLM post-training framework not specifically focused on software engineering or code.
Benchmarking Embedding Geometries and Clustering Algorithms for Unsupervised Patent Claim Classification Compares hyperbolic and Euclidean embeddings for unsupervised patent representation to improve cross-category prior-art discovery. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Does not address safety, jailbreaking, red-teaming, or model guardrails.
Benchmarking Embedding Geometries and Clustering Algorithms for Unsupervised Patent Claim Classification Compares hyperbolic and Euclidean embeddings for unsupervised patent representation to improve cross-category prior-art discovery. Preprint	Small and efficient models via arxiv · 12 Oct, 16:28	Not relevant Examines representation geometry rather than model compression, quantization, or efficient serving.
Benchmarking Embedding Geometries and Clustering Algorithms for Unsupervised Patent Claim Classification Compares hyperbolic and Euclidean embeddings for unsupervised patent representation to improve cross-category prior-art discovery. Preprint	Multimodal models via arxiv · 12 Oct, 16:28	Not relevant Operates exclusively on patent text without multimodal image or audio modalities.
Benchmarking Embedding Geometries and Clustering Algorithms for Unsupervised Patent Claim Classification Compares hyperbolic and Euclidean embeddings for unsupervised patent representation to improve cross-category prior-art discovery. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Evaluates clustering representations, not language model benchmarking or LLM-as-a-judge.
Benchmarking Embedding Geometries and Clustering Algorithms for Unsupervised Patent Claim Classification Compares hyperbolic and Euclidean embeddings for unsupervised patent representation to improve cross-category prior-art discovery. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Study of embedding geometry and clustering for patent search, not multi-step tool-using agents.
Benchmarking Embedding Geometries and Clustering Algorithms for Unsupervised Patent Claim Classification Compares hyperbolic and Euclidean embeddings for unsupervised patent representation to improve cross-category prior-art discovery. Preprint	LLMs for code via arxiv · 12 Oct, 16:28	Not relevant Applies to patent claims rather than software code or programming assistants.
Hindsight Structured Advice Distillation: Steering Executor Agents with Advisors Trained on Their Failures and Task Characteristics Trains an advisor model on past failures to add dynamic, structured guidance to frozen language model agent prompts. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Focuses on an agent advisor training framework rather than LLM benchmarking or evaluation methodology.
Hindsight Structured Advice Distillation: Steering Executor Agents with Advisors Trained on Their Failures and Task Characteristics Trains an advisor model on past failures to add dynamic, structured guidance to frozen language model agent prompts. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Deals with domain-specific agent guidance rather than safety, guardrails, or alignment.
Scaling Empowered Agents Presents a scalable method to maximize agent empowerment by guiding agents toward states affording numerous executable skills. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Explores unsupervised RL objectives rather than LLM benchmarks or evaluation methodologies.
Scaling Empowered Agents Presents a scalable method to maximize agent empowerment by guiding agents toward states affording numerous executable skills. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Studies RL empowerment and skill learning in environments like Crafter, not LLM agent systems.
Scaling Empowered Agents Presents a scalable method to maximize agent empowerment by guiding agents toward states affording numerous executable skills. Preprint	Small and efficient models via arxiv · 12 Oct, 16:28	Not relevant Investigates reinforcement learning empowerment scaling, not model compression or efficient serving.
Scaling Empowered Agents Presents a scalable method to maximize agent empowerment by guiding agents toward states affording numerous executable skills. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Focuses on intrinsic motivation in RL rather than AI safety, red-teaming, or guardrails.
Provenance-Aware Belief Updating in Human-AI Feedback Loops Introduces a training objective enabling language models to update beliefs using report lineage rather than raw citation repetition. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Investigates Bayesian belief updates from relayed inputs, not multi-step autonomous agents.
Provenance-Aware Belief Updating in Human-AI Feedback Loops Introduces a training objective enabling language models to update beliefs using report lineage rather than raw citation repetition. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Proposes a training method for belief updating rather than a comprehensive evaluation benchmark.
Generalized Proximal Policy Optimization: Unifying LLM RL Algorithms from the Perspective of Divergence Proposes a generalized f-divergence trust-region framework to unify and stabilize reinforcement learning algorithms for LLM post-training. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Presents a theoretical RL optimization framework, not autonomous multi-step agents or tool use.
Generalized Proximal Policy Optimization: Unifying LLM RL Algorithms from the Perspective of Divergence Proposes a generalized f-divergence trust-region framework to unify and stabilize reinforcement learning algorithms for LLM post-training. Preprint	Small and efficient models via arxiv · 12 Oct, 16:28	Not relevant Addresses RL algorithmic formulation rather than model compression, quantization, or serving.
Generalized Proximal Policy Optimization: Unifying LLM RL Algorithms from the Perspective of Divergence Proposes a generalized f-divergence trust-region framework to unify and stabilize reinforcement learning algorithms for LLM post-training. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Focuses on reinforcement learning policy optimization rather than benchmark design or evaluation.
Generalized Proximal Policy Optimization: Unifying LLM RL Algorithms from the Perspective of Divergence Proposes a generalized f-divergence trust-region framework to unify and stabilize reinforcement learning algorithms for LLM post-training. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Focuses on general RL mathematical theory and convergence rather than safety alignment or guardrails.
EmbodiedCLBench: Evaluating Continual Learning for Self-Evolving Embodied Agents Presents a simulation benchmark evaluating how embodied agents continually acquire basic skills and recombine them for complex tasks. Preprint	Multimodal models via arxiv · 12 Oct, 16:28	Not relevant Focuses on continual learning benchmark design rather than multimodal modeling architectures.
EmbodiedCLBench: Evaluating Continual Learning for Self-Evolving Embodied Agents Presents a simulation benchmark evaluating how embodied agents continually acquire basic skills and recombine them for complex tasks. Preprint	LLMs for code via arxiv · 12 Oct, 16:28	Not relevant Evaluates embodied task execution in simulation, not software engineering or code generation.
EmbodiedCLBench: Evaluating Continual Learning for Self-Evolving Embodied Agents Presents a simulation benchmark evaluating how embodied agents continually acquire basic skills and recombine them for complex tasks. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Focuses on skill generalization in embodied settings rather than AI safety or reliability.
EmbodiedCLBench: Evaluating Continual Learning for Self-Evolving Embodied Agents Presents a simulation benchmark evaluating how embodied agents continually acquire basic skills and recombine them for complex tasks. Preprint	Small and efficient models via arxiv · 12 Oct, 16:28	Not relevant Addresses continual learning benchmarks rather than model compression or efficient inference.
EmbodiedCLBench: Evaluating Continual Learning for Self-Evolving Embodied Agents Presents a simulation benchmark evaluating how embodied agents continually acquire basic skills and recombine them for complex tasks. Preprint	Computer vision for industrial applications via arxiv · 12 Oct, 16:28	Not relevant Focuses on embodied AI simulation rather than industrial visual inspection or manufacturing.
EmbodiedCLBench: Evaluating Continual Learning for Self-Evolving Embodied Agents Presents a simulation benchmark evaluating how embodied agents continually acquire basic skills and recombine them for complex tasks. Preprint	Retrieval-augmented generation (RAG) via arxiv · 12 Oct, 16:28	Not relevant Focuses on embodied skill acquisition and transfer, not document retrieval or RAG pipelines.
Don't Guess, Just Ask: Evaluating Interactive Requirement Clarification for Repository-Level Software Development Introduces a benchmark to evaluate how effectively LLMs clarify ambiguous user requirements in repository-level software development tasks. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Focuses on requirement clarification benchmarks rather than multi-step autonomous agent architectures.
Don't Guess, Just Ask: Evaluating Interactive Requirement Clarification for Repository-Level Software Development Introduces a benchmark to evaluate how effectively LLMs clarify ambiguous user requirements in repository-level software development tasks. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Addresses requirement ambiguity in software engineering rather than safety, alignment, or guardrails.
GeoGR: Spatial-Semantic Token Compression and Group Routing for Efficient 3D VLMs Proposes spatial-semantic token compression and layer routing to accelerate 3D vision-language model inference across multi-view inputs. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Proposes an architectural efficiency framework rather than an evaluation benchmark.
GeoGR: Spatial-Semantic Token Compression and Group Routing for Efficient 3D VLMs Proposes spatial-semantic token compression and layer routing to accelerate 3D vision-language model inference across multi-view inputs. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Addresses 3D vision-language representation compression, not autonomous multi-step agents.
GeoGR: Spatial-Semantic Token Compression and Group Routing for Efficient 3D VLMs Proposes spatial-semantic token compression and layer routing to accelerate 3D vision-language model inference across multi-view inputs. Preprint	LLMs for code via arxiv · 12 Oct, 16:28	Not relevant Does not involve software development, code generation, or program analysis.
GeoGR: Spatial-Semantic Token Compression and Group Routing for Efficient 3D VLMs Proposes spatial-semantic token compression and layer routing to accelerate 3D vision-language model inference across multi-view inputs. Preprint	Retrieval-augmented generation (RAG) via arxiv · 12 Oct, 16:28	Not relevant Focuses on 3D token compression during inference, not document retrieval or grounding.
GeoGR: Spatial-Semantic Token Compression and Group Routing for Efficient 3D VLMs Proposes spatial-semantic token compression and layer routing to accelerate 3D vision-language model inference across multi-view inputs. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Focuses on computational efficiency rather than model safety, reliability, or guardrails.
GeoGR: Spatial-Semantic Token Compression and Group Routing for Efficient 3D VLMs Proposes spatial-semantic token compression and layer routing to accelerate 3D vision-language model inference across multi-view inputs. Preprint	Computer vision for industrial applications via arxiv · 12 Oct, 16:28	Not relevant Focuses on general 3D VLM inference rather than industrial inspection or manufacturing perception.
CoBe: A Benchmark For Conversational Counterfactual Text Editing Presents a conversational benchmark testing whether models can edit text under counterfactual changes while preserving unaffected factual details. Preprint	Retrieval-augmented generation (RAG) via arxiv · 12 Oct, 16:28	Not relevant Addresses counterfactual causal reasoning in text, not document retrieval or knowledge grounding.
CoBe: A Benchmark For Conversational Counterfactual Text Editing Presents a conversational benchmark testing whether models can edit text under counterfactual changes while preserving unaffected factual details. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Evaluates causal text modification rather than safety guardrails, jailbreaks, or deployment risks.
CoBe: A Benchmark For Conversational Counterfactual Text Editing Presents a conversational benchmark testing whether models can edit text under counterfactual changes while preserving unaffected factual details. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Assesses counterfactual text rewriting models, not multi-step autonomous agents or tool use.
CoBe: A Benchmark For Conversational Counterfactual Text Editing Presents a conversational benchmark testing whether models can edit text under counterfactual changes while preserving unaffected factual details. Preprint	Multimodal models via arxiv · 12 Oct, 16:28	Not relevant Focuses entirely on text editing and natural language counterfactuals, with no multimodal aspects.
CoBe: A Benchmark For Conversational Counterfactual Text Editing Presents a conversational benchmark testing whether models can edit text under counterfactual changes while preserving unaffected factual details. Preprint	LLMs for code via arxiv · 12 Oct, 16:28	Not relevant Focuses on conversational natural language editing rather than code generation or repair.
CoBe: A Benchmark For Conversational Counterfactual Text Editing Presents a conversational benchmark testing whether models can edit text under counterfactual changes while preserving unaffected factual details. Preprint	Small and efficient models via arxiv · 12 Oct, 16:28	Not relevant Does not address model compression, distillation, quantization, or inference acceleration.
Travel Back To Move Forward: Retrospective Exploration and Evolving for Long Horizon Agents Introduces a framework allowing agents to rewind long-horizon environments to earlier states while retaining learned exploratory knowledge. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Focuses on task-completion search policies rather than safety guardrails or adversarial robustness.
Travel Back To Move Forward: Retrospective Exploration and Evolving for Long Horizon Agents Introduces a framework allowing agents to rewind long-horizon environments to earlier states while retaining learned exploratory knowledge. Preprint	Retrieval-augmented generation (RAG) via arxiv · 12 Oct, 16:28	Not relevant Deals with reinforcement learning and state rollbacks, not document retrieval or grounding.
Travel Back To Move Forward: Retrospective Exploration and Evolving for Long Horizon Agents Introduces a framework allowing agents to rewind long-horizon environments to earlier states while retaining learned exploratory knowledge. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Uses established benchmarks to assess agent backtracking rather than proposing new evaluations.
Travel Back To Move Forward: Retrospective Exploration and Evolving for Long Horizon Agents Introduces a framework allowing agents to rewind long-horizon environments to earlier states while retaining learned exploratory knowledge. Preprint	Multimodal models via arxiv · 12 Oct, 16:28	Not relevant Focuses on decision-making policies in interactive text/grid environments, not multimodal models.
Travel Back To Move Forward: Retrospective Exploration and Evolving for Long Horizon Agents Introduces a framework allowing agents to rewind long-horizon environments to earlier states while retaining learned exploratory knowledge. Preprint	LLMs for code via arxiv · 12 Oct, 16:28	Not relevant Evaluates on interactive environments like Sokoban and ALFWorld, not software engineering or coding.
Travel Back To Move Forward: Retrospective Exploration and Evolving for Long Horizon Agents Introduces a framework allowing agents to rewind long-horizon environments to earlier states while retaining learned exploratory knowledge. Preprint	Small and efficient models via arxiv · 12 Oct, 16:28	Not relevant Focuses on exploration policies and state rewinding, not model compression or efficient serving.
Evolving Creatives with Clicks: CTR-Driven Advertising image Editing via Attribute Disentanglement and Preference Alignment Proposes an advertising image editing method that identifies and selectively refines visual attributes driving click-through rates. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Focuses on CTR-driven image editing rather than autonomous planning or agent tool use.
Evolving Creatives with Clicks: CTR-Driven Advertising image Editing via Attribute Disentanglement and Preference Alignment Proposes an advertising image editing method that identifies and selectively refines visual attributes driving click-through rates. Preprint	Computer vision for industrial applications via arxiv · 12 Oct, 16:28	Not relevant Deals with e-commerce ad creative editing, not industrial manufacturing or robotic vision.
Evolving Creatives with Clicks: CTR-Driven Advertising image Editing via Attribute Disentanglement and Preference Alignment Proposes an advertising image editing method that identifies and selectively refines visual attributes driving click-through rates. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Proposes an advertising image optimization pipeline, not an LLM or vision evaluation benchmark.
Evolving Creatives with Clicks: CTR-Driven Advertising image Editing via Attribute Disentanglement and Preference Alignment Proposes an advertising image editing method that identifies and selectively refines visual attributes driving click-through rates. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Optimizes marketing conversion metrics rather than model safety, security, or robustness.
LongHarness: Advancing Agentic Reasoning over 100M-Token Corpora via Stateful Evidence Management Presents an agent harness that coevolves evidence and conclusions to improve reasoning across 100-million-token document corpora. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Focuses on information retrieval and reasoning state tracking rather than safety or guardrails.
LongHarness: Advancing Agentic Reasoning over 100M-Token Corpora via Stateful Evidence Management Presents an agent harness that coevolves evidence and conclusions to improve reasoning across 100-million-token document corpora. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Introduces an agent framework for corpus-level reasoning, not a general evaluation benchmark.
Attention Sees, MLP Refuses: Tracing the Attention-to-MLP Safety Pathway in LLMs Investigates internal transformer mechanisms, showing attention updates decode harmfulness while MLP layers primarily drive refusal token generation. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Investigates internal transformer layer mechanics, not autonomous multi-step agents or tool use.
Attention Sees, MLP Refuses: Tracing the Attention-to-MLP Safety Pathway in LLMs Investigates internal transformer mechanisms, showing attention updates decode harmfulness while MLP layers primarily drive refusal token generation. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Focuses on mechanistic interpretability of safety mechanisms rather than benchmarking LLM systems.
SWEAR: Section-Wise Efficient Audio Reasoning via Self-Rated Difficulty Presents a difficulty-aware framework that allocates reasoning steps across sections to improve efficiency in large audio-language models. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Focuses on an adaptive inference framework evaluated on standard benchmarks, not evaluation design.
SWEAR: Section-Wise Efficient Audio Reasoning via Self-Rated Difficulty Presents a difficulty-aware framework that allocates reasoning steps across sections to improve efficiency in large audio-language models. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Addresses inference efficiency and token allocation rather than safety, guardrails, or alignment.
SWEAR: Section-Wise Efficient Audio Reasoning via Self-Rated Difficulty Presents a difficulty-aware framework that allocates reasoning steps across sections to improve efficiency in large audio-language models. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Examines internal token-budget allocation for audio reasoning, not agent planning or tool use.
VETR: Execution-Grounded Self-Improving Agent for Symbolic Graphics Reasoning Introduces a recursive self-improvement agent framework that refines symbolic graphics programs using execution traces and runtime state changes. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Evaluates on existing Blender benchmarks rather than introducing a new evaluation suite.
VETR: Execution-Grounded Self-Improving Agent for Symbolic Graphics Reasoning Introduces a recursive self-improvement agent framework that refines symbolic graphics programs using execution traces and runtime state changes. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Focuses on symbolic graphics reasoning and refinement rather than deployment safety or guardrails.
Does Location Matter? A Black-Box Causal Audit Reveals Model-Dependent Audio Grounding in LALM Hallucination Audits audio-language model hallucination mitigations by testing localized acoustic evidence ablation against baseline methods across model architectures. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Studies multimodal audio-language decoding and hallucination, not tool-using or autonomous agent systems.

13,441–13,500 of 13,754← Previous Next →