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
Memory as Weights: Internalizing Long-Term History for Streaming Videos The paper compresses video history into evolving lightweight model weights instead of competing for limited input context length. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:17	Not relevant Addresses streaming video-language models, not industrial defect detection or manufacturing inspection.
Memory as Weights: Internalizing Long-Term History for Streaming Videos The paper compresses video history into evolving lightweight model weights instead of competing for limited input context length. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Does not address safety, jailbreaks, robustness, or deployment monitoring.
Memory as Weights: Internalizing Long-Term History for Streaming Videos The paper compresses video history into evolving lightweight model weights instead of competing for limited input context length. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Focuses on streaming video memory, not tool use or autonomous agents.
Learning from Unverified Peers via Shared Residual Distillation The paper develops a self-distillation approach enabling models to learn from their own unverified alternative reasoning paths without ground-truth answers. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Studies self-distillation training methods, not autonomous tool-use or multi-step agents.
Learning from Unverified Peers via Shared Residual Distillation The paper develops a self-distillation approach enabling models to learn from their own unverified alternative reasoning paths without ground-truth answers. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Addresses training via self-distillation rather than safety, red-teaming, or guardrails.
Learning from Unverified Peers via Shared Residual Distillation The paper develops a self-distillation approach enabling models to learn from their own unverified alternative reasoning paths without ground-truth answers. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Introduces a distillation training technique rather than testing or benchmarking methodologies.
Multi-SFT: Permissively-Licensed Multilingual Instruction-Following The paper introduces a commercially permissive multilingual instruction dataset covering five European languages to train open language models. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Creates a general instruction-tuning corpus rather than dedicated safety guardrails or alignment frameworks.
Multi-SFT: Permissively-Licensed Multilingual Instruction-Following The paper introduces a commercially permissive multilingual instruction dataset covering five European languages to train open language models. Preprint	Small and efficient models via arxiv · 17 Oct, 06:17	Not relevant Focuses on instruction data creation rather than model compression, quantization, or distillation.
Multi-SFT: Permissively-Licensed Multilingual Instruction-Following The paper introduces a commercially permissive multilingual instruction dataset covering five European languages to train open language models. Preprint	Retrieval-augmented generation (RAG) via arxiv · 17 Oct, 06:17	Not relevant Focuses on instruction-tuning datasets, not retrieval-augmented generation or external document grounding.
Multi-SFT: Permissively-Licensed Multilingual Instruction-Following The paper introduces a commercially permissive multilingual instruction dataset covering five European languages to train open language models. Preprint	LLMs for code via arxiv · 17 Oct, 06:17	Not relevant Covers general multilingual instruction following, not software engineering or code-specific LLM tasks.
Multi-SFT: Permissively-Licensed Multilingual Instruction-Following The paper introduces a commercially permissive multilingual instruction dataset covering five European languages to train open language models. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Presents an instruction dataset, not multi-step tool-using autonomous agent architectures.
Multi-SFT: Permissively-Licensed Multilingual Instruction-Following The paper introduces a commercially permissive multilingual instruction dataset covering five European languages to train open language models. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Contributes an instruction-tuning dataset rather than proposing new evaluation methodologies or benchmarks.
Multi-SFT: Permissively-Licensed Multilingual Instruction-Following The paper introduces a commercially permissive multilingual instruction dataset covering five European languages to train open language models. Preprint	Multimodal models via arxiv · 17 Oct, 06:17	Not relevant Exclusively focuses on text instruction datasets across European languages, not multimodality.
TIPS: A Tiny Inference-time Policy for Scaling in Diffusion Language Models The paper presents a tiny learned policy controller to optimize token unmasking and remasking schedules in diffusion language models. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Focuses on decoding algorithms rather than novel evaluation methodologies or benchmarks.
TIPS: A Tiny Inference-time Policy for Scaling in Diffusion Language Models The paper presents a tiny learned policy controller to optimize token unmasking and remasking schedules in diffusion language models. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Does not address AI safety, jailbreaks, hallucinations, or deployment guardrails.
TIPS: A Tiny Inference-time Policy for Scaling in Diffusion Language Models The paper presents a tiny learned policy controller to optimize token unmasking and remasking schedules in diffusion language models. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Proposes a diffusion decoding policy, not tool use or LLM agents.
Causal Integrated Gradients: Scalable Feature Attribution Along Causal Paths The paper extends integrated gradients to trace feature attributions along causal paths for high-dimensional models. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Evaluates explainability methods, not LLM capabilities, judges, or evaluation benchmarks.
Causal Integrated Gradients: Scalable Feature Attribution Along Causal Paths The paper extends integrated gradients to trace feature attributions along causal paths for high-dimensional models. Preprint	Small and efficient models via arxiv · 17 Oct, 06:17	Not relevant Does not address model compression, quantization, distillation, or efficient model serving.
Causal Integrated Gradients: Scalable Feature Attribution Along Causal Paths The paper extends integrated gradients to trace feature attributions along causal paths for high-dimensional models. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Addresses explainability methods rather than deployment safety, jailbreaks, or guardrails.
Causal Integrated Gradients: Scalable Feature Attribution Along Causal Paths The paper extends integrated gradients to trace feature attributions along causal paths for high-dimensional models. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Focuses on feature attribution, not multi-step LLM planning or agentic workflows.
Explaining the Roles of Experts in Mixture-of-Experts LLMs for Personalization The paper personalizes mixture-of-experts models and analyzes which specialized experts encode distinct user-specific features. Preprint	Multimodal models via arxiv · 17 Oct, 06:17	Not relevant Investigates text-only MoE models rather than multimodal capabilities.
Explaining the Roles of Experts in Mixture-of-Experts LLMs for Personalization The paper personalizes mixture-of-experts models and analyzes which specialized experts encode distinct user-specific features. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Focuses on mechanistic interpretability rather than evaluation benchmarks.
Explaining the Roles of Experts in Mixture-of-Experts LLMs for Personalization The paper personalizes mixture-of-experts models and analyzes which specialized experts encode distinct user-specific features. Preprint	LLMs for code via arxiv · 17 Oct, 06:17	Not relevant Focuses on personalization and interpretability, not software code LLMs.
Explaining the Roles of Experts in Mixture-of-Experts LLMs for Personalization The paper personalizes mixture-of-experts models and analyzes which specialized experts encode distinct user-specific features. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Analyzes MoE routing for personalization, not multi-step agents.
Explaining the Roles of Experts in Mixture-of-Experts LLMs for Personalization The paper personalizes mixture-of-experts models and analyzes which specialized experts encode distinct user-specific features. Preprint	Retrieval-augmented generation (RAG) via arxiv · 17 Oct, 06:17	Not relevant Examines internal MoE representations, not external retrieval systems.
Explaining the Roles of Experts in Mixture-of-Experts LLMs for Personalization The paper personalizes mixture-of-experts models and analyzes which specialized experts encode distinct user-specific features. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Studies MoE personalization interpretability, not safety or guardrails.
DropKV: Decoupling Residual-Output Perturbation for Near-Optimal KV-Cache Eviction The paper proposes a decoupled per-token scoring method to efficiently manage key-value cache memory during long-context generation. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Covers inference speed and memory, not safety or guardrails.
DropKV: Decoupling Residual-Output Perturbation for Near-Optimal KV-Cache Eviction The paper proposes a decoupled per-token scoring method to efficiently manage key-value cache memory during long-context generation. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Proposes an eviction algorithm, not an evaluation benchmark.
DropKV: Decoupling Residual-Output Perturbation for Near-Optimal KV-Cache Eviction The paper proposes a decoupled per-token scoring method to efficiently manage key-value cache memory during long-context generation. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Focuses on KV-cache management, not autonomous multi-step agents.
CART: Compatible Routing for Multimodal Continual Instruction Tuning The paper coordinates projector and language model adaptation using task-specific routing to reduce catastrophic forgetting in multimodal models. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Proposes a training method, not an evaluation benchmark.
CART: Compatible Routing for Multimodal Continual Instruction Tuning The paper coordinates projector and language model adaptation using task-specific routing to reduce catastrophic forgetting in multimodal models. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Focuses on multimodal continual tuning, not autonomous AI agents.
CART: Compatible Routing for Multimodal Continual Instruction Tuning The paper coordinates projector and language model adaptation using task-specific routing to reduce catastrophic forgetting in multimodal models. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Focuses on continual learning drift, not deployment guardrails or safety.
Believed but Not Said: Measuring Hidden Belief Collapse in Multi-Agent LLM Discussions The paper demonstrates that a single adversarial agent can cause benign models in multi-agent discussions to quietly adopt false beliefs. Preprint	Multimodal models via arxiv · 17 Oct, 06:17	Not relevant Focuses on text-based multi-agent discussions without multimodal inputs.
Believed but Not Said: Measuring Hidden Belief Collapse in Multi-Agent LLM Discussions The paper demonstrates that a single adversarial agent can cause benign models in multi-agent discussions to quietly adopt false beliefs. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Investigates adversarial peer dynamics, not evaluation benchmark development.
Believed but Not Said: Measuring Hidden Belief Collapse in Multi-Agent LLM Discussions The paper demonstrates that a single adversarial agent can cause benign models in multi-agent discussions to quietly adopt false beliefs. Preprint	Retrieval-augmented generation (RAG) via arxiv · 17 Oct, 06:17	Not relevant Studies multi-agent discussion robustness, not retrieval or RAG systems.
Believed but Not Said: Measuring Hidden Belief Collapse in Multi-Agent LLM Discussions The paper demonstrates that a single adversarial agent can cause benign models in multi-agent discussions to quietly adopt false beliefs. Preprint	LLMs for code via arxiv · 17 Oct, 06:17	Not relevant Studies multi-agent factual QA, not code generation or repair.
Believed but Not Said: Measuring Hidden Belief Collapse in Multi-Agent LLM Discussions The paper demonstrates that a single adversarial agent can cause benign models in multi-agent discussions to quietly adopt false beliefs. Preprint	Small and efficient models via arxiv · 17 Oct, 06:17	Not relevant Examines multi-agent consensus dynamics, not model efficiency or distillation.
Are You Sure Your LLM Is Sure? Repairing and Auditing Confidence The paper introduces a framework to recalibrate, audit, and safely deploy confidence scores for language model outputs. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Focuses on model confidence scoring, not multi-step agents.
RAG-RST: Retrieved Exemplars as In-Context Discourse Schema for RST Parsing The paper uses retrieved in-context exemplars to improve rhetorical structure discourse parsing without requiring model fine-tuning. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Applies retrieval to parsing, not autonomous agent planning.
RAG-RST: Retrieved Exemplars as In-Context Discourse Schema for RST Parsing The paper uses retrieved in-context exemplars to improve rhetorical structure discourse parsing without requiring model fine-tuning. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Addresses discourse parsing, not safety, jailbreaks, or guardrails.
RAG-RST: Retrieved Exemplars as In-Context Discourse Schema for RST Parsing The paper uses retrieved in-context exemplars to improve rhetorical structure discourse parsing without requiring model fine-tuning. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Presents a parsing method, not an LLM benchmark framework.
Specification Gaming and a Constrained Reinforcement Learning Mitigation in a Cleaning Gridworld The paper demonstrates specification gaming in a simulated cleaning task and tests a constrained reinforcement learning mitigation. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Focuses on toy gridworld RL reward gaming, not LLM agents.
Specification Gaming and a Constrained Reinforcement Learning Mitigation in a Cleaning Gridworld The paper demonstrates specification gaming in a simulated cleaning task and tests a constrained reinforcement learning mitigation. Preprint	Small and efficient models via arxiv · 17 Oct, 06:17	Not relevant Deals with reward hacking in RL, not model efficiency.
Specification Gaming and a Constrained Reinforcement Learning Mitigation in a Cleaning Gridworld The paper demonstrates specification gaming in a simulated cleaning task and tests a constrained reinforcement learning mitigation. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Analyzes specification gaming in gridworld RL, not LLM benchmarks.
Realtime-Bench: Can Frontier Models Solve Real-time Interactive Problems? The paper introduces a benchmark to measure language model agent performance under real-time latency and streaming temporal constraints. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Studies real-time latency evaluation, not AI safety or guardrails.
EMO: A Closed-Loop Test-Time Computation Framework via Alternating Context Optimization and Logits Steering The paper proposes an alternating context optimization and logit steering framework to adapt language model outputs at test time. Preprint	Retrieval-augmented generation (RAG) via arxiv · 17 Oct, 06:17	Not relevant Focuses on context optimization and logit steering, not RAG.
EMO: A Closed-Loop Test-Time Computation Framework via Alternating Context Optimization and Logits Steering The paper proposes an alternating context optimization and logit steering framework to adapt language model outputs at test time. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Studies test-time context optimization, not AI safety or guardrails.
EMO: A Closed-Loop Test-Time Computation Framework via Alternating Context Optimization and Logits Steering The paper proposes an alternating context optimization and logit steering framework to adapt language model outputs at test time. Preprint	Multimodal models via arxiv · 17 Oct, 06:17	Not relevant Operates strictly on text generation rather than multimodal architectures.
EMO: A Closed-Loop Test-Time Computation Framework via Alternating Context Optimization and Logits Steering The paper proposes an alternating context optimization and logit steering framework to adapt language model outputs at test time. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Proposes an inference adaptation method, not an evaluation benchmark.
EMO: A Closed-Loop Test-Time Computation Framework via Alternating Context Optimization and Logits Steering The paper proposes an alternating context optimization and logit steering framework to adapt language model outputs at test time. Preprint	LLMs for code via arxiv · 17 Oct, 06:17	Not relevant Focuses on test-time optimization theory, not code generation.
EMO: A Closed-Loop Test-Time Computation Framework via Alternating Context Optimization and Logits Steering The paper proposes an alternating context optimization and logit steering framework to adapt language model outputs at test time. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Addresses test-time logits steering, not multi-step tool-using agents.
Multi-sourced AI Tutor for Non-Native English Speaker Students The paper develops a multilingual retrieval-augmented tutoring system providing bilingual responses and glossaries for international computing students. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant An educational QA system, not an autonomous planning agent.
Multi-sourced AI Tutor for Non-Native English Speaker Students The paper develops a multilingual retrieval-augmented tutoring system providing bilingual responses and glossaries for international computing students. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Discusses an educational chatbot, not AI safety or guardrails.
Multi-sourced AI Tutor for Non-Native English Speaker Students The paper develops a multilingual retrieval-augmented tutoring system providing bilingual responses and glossaries for international computing students. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Focuses on educational QA application, not LLM benchmark design.
Mosaic: Multi-Session Agentic Reinforcement Learning The paper introduces a hierarchical reinforcement learning method designed for agentic tasks that span multiple interaction sessions. Preprint	Small and efficient models via arxiv · 17 Oct, 06:17	Not relevant Studies agent training dynamics rather than model compression techniques.
Mosaic: Multi-Session Agentic Reinforcement Learning The paper introduces a hierarchical reinforcement learning method designed for agentic tasks that span multiple interaction sessions. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Proposes an RL training algorithm, not an evaluation benchmark.
Mosaic: Multi-Session Agentic Reinforcement Learning The paper introduces a hierarchical reinforcement learning method designed for agentic tasks that span multiple interaction sessions. Preprint	LLMs for code via arxiv · 17 Oct, 06:17	Not relevant Investigates general agent trajectory optimization, not code generation.
Mosaic: Multi-Session Agentic Reinforcement Learning The paper introduces a hierarchical reinforcement learning method designed for agentic tasks that span multiple interaction sessions. Preprint	Retrieval-augmented generation (RAG) via arxiv · 17 Oct, 06:17	Not relevant Focuses on multi-session reinforcement learning, not retrieval systems.
Mosaic: Multi-Session Agentic Reinforcement Learning The paper introduces a hierarchical reinforcement learning method designed for agentic tasks that span multiple interaction sessions. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Addresses reinforcement learning optimization, not deployment safety or guardrails.
Mosaic: Multi-Session Agentic Reinforcement Learning The paper introduces a hierarchical reinforcement learning method designed for agentic tasks that span multiple interaction sessions. Preprint	Multimodal models via arxiv · 17 Oct, 06:17	Not relevant Focuses on text agent trajectories without multimodal processing.

5,581–5,640 of 17,313← Previous Next →