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
Learning to Read Out: Unembedding Dynamics in Language Model Pretraining The paper investigates when language models learn capabilities by tracking unembedding matrix dynamics during pretraining using parameter-trajectory crosscoding. Preprint	LLMs for code via arxiv · 17 Oct, 06:17	Not relevant Focuses on pretraining unembedding dynamics, not software code tasks.
Learning to Read Out: Unembedding Dynamics in Language Model Pretraining The paper investigates when language models learn capabilities by tracking unembedding matrix dynamics during pretraining using parameter-trajectory crosscoding. Preprint	Multimodal models via arxiv · 17 Oct, 06:17	Not relevant Focuses on text language model pretraining, not multimodal inputs.
Learning to Read Out: Unembedding Dynamics in Language Model Pretraining The paper investigates when language models learn capabilities by tracking unembedding matrix dynamics during pretraining using parameter-trajectory crosscoding. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Investigates representation probing during pretraining rather than LLM benchmarks.
Learning to Read Out: Unembedding Dynamics in Language Model Pretraining The paper investigates when language models learn capabilities by tracking unembedding matrix dynamics during pretraining using parameter-trajectory crosscoding. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Studies internal representation dynamics rather than safety or guardrails.
Learning to Read Out: Unembedding Dynamics in Language Model Pretraining The paper investigates when language models learn capabilities by tracking unembedding matrix dynamics during pretraining using parameter-trajectory crosscoding. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Analyzes pretraining dynamics rather than multi-step tool-using agents.
SCARCE: Emergent Institutions in Long-Horizon Multi-Agent Coding Systems The paper studies how autonomous coding agents spontaneously negotiate and coordinate access to shared compute resources under deadlines. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Studies multi-agent emergent behavior rather than standard LLM evaluation benchmarks.
Recursive Reward Improvement via Adversarial Meta-Prompt Evolution Proposes an adversarial meta-prompt evolution method to improve model-generated evaluation rubrics for open-ended tasks. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Addresses RL training optimization rather than benchmark methodologies.
Recursive Reward Improvement via Adversarial Meta-Prompt Evolution Proposes an adversarial meta-prompt evolution method to improve model-generated evaluation rubrics for open-ended tasks. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Focuses on RL training loops and adversarial rubrics rather than interactive agents.
ReviewPilot: A Multi-Agent System for Screening and Data Extraction in Systematic Reviews Presents a multi-agent system for paper screening and data extraction in systematic literature reviews following standard guidelines. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Studies literature review automation rather than AI safety or alignment.
ReviewPilot: A Multi-Agent System for Screening and Data Extraction in Systematic Reviews Presents a multi-agent system for paper screening and data extraction in systematic literature reviews following standard guidelines. Preprint	LLMs for code via arxiv · 17 Oct, 06:17	Not relevant Applies to systematic biomedical literature reviews, not software engineering.
ReviewPilot: A Multi-Agent System for Screening and Data Extraction in Systematic Reviews Presents a multi-agent system for paper screening and data extraction in systematic literature reviews following standard guidelines. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Presents an application system rather than an evaluation benchmark.
ReviewPilot: A Multi-Agent System for Screening and Data Extraction in Systematic Reviews Presents a multi-agent system for paper screening and data extraction in systematic literature reviews following standard guidelines. Preprint	Small and efficient models via arxiv · 17 Oct, 06:17	Not relevant Does not study model compression, distillation, or efficient serving.
ReviewPilot: A Multi-Agent System for Screening and Data Extraction in Systematic Reviews Presents a multi-agent system for paper screening and data extraction in systematic literature reviews following standard guidelines. Preprint	Multimodal models via arxiv · 17 Oct, 06:17	Not relevant Processes text papers rather than multimodal or vision inputs.
Learning When to Withdraw: Curriculum Coverage and Rule-Specific Failures in Fine-Tuned Language Agents Studies how fine-tuning helps tool-using agents learn to withdraw and repair actions when environment assumptions change. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Studies agent action withdrawal training rather than proposing a general benchmark.
PremiseCheck: Efficient Fact-Checking of Human and AI Peer Reviews Combines language models and lightweight decision models to fact-check statements in peer reviews against original manuscripts. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Focuses on peer-review fact-checking pipelines rather than multi-step tool-using agents.
PremiseCheck: Efficient Fact-Checking of Human and AI Peer Reviews Combines language models and lightweight decision models to fact-check statements in peer reviews against original manuscripts. Preprint	LLMs for code via arxiv · 17 Oct, 06:17	Not relevant Deals with natural language peer reviews rather than software code.
PremiseCheck: Efficient Fact-Checking of Human and AI Peer Reviews Combines language models and lightweight decision models to fact-check statements in peer reviews against original manuscripts. Preprint	Small and efficient models via arxiv · 17 Oct, 06:17	Not relevant Combines models for cost efficiency but does not focus on small model techniques.
PremiseCheck: Efficient Fact-Checking of Human and AI Peer Reviews Combines language models and lightweight decision models to fact-check statements in peer reviews against original manuscripts. Preprint	Multimodal models via arxiv · 17 Oct, 06:17	Not relevant Focuses strictly on text documents without multimodal inputs.
PremiseCheck: Efficient Fact-Checking of Human and AI Peer Reviews Combines language models and lightweight decision models to fact-check statements in peer reviews against original manuscripts. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Focuses on factual accuracy in reviews rather than safety or alignment guardrails.
Athanor: An Agent-Native System for Autonomous Post-Training Research Develops an open agent environment for evaluating how autonomously language agents can conduct post-training machine learning experiments. Preprint	Multimodal models via arxiv · 17 Oct, 06:17	Not relevant Does not evaluate or build multimodal vision-language models.
Athanor: An Agent-Native System for Autonomous Post-Training Research Develops an open agent environment for evaluating how autonomously language agents can conduct post-training machine learning experiments. Preprint	Retrieval-augmented generation (RAG) via arxiv · 17 Oct, 06:17	Not relevant Does not focus on retrieval-augmented generation architectures.
Athanor: An Agent-Native System for Autonomous Post-Training Research Develops an open agent environment for evaluating how autonomously language agents can conduct post-training machine learning experiments. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Introduces an agent system for post-training rather than an evaluation benchmark.
Athanor: An Agent-Native System for Autonomous Post-Training Research Develops an open agent environment for evaluating how autonomously language agents can conduct post-training machine learning experiments. Preprint	Small and efficient models via arxiv · 17 Oct, 06:17	Not relevant Uses small base models as subjects but does not focus on efficiency techniques.
Athanor: An Agent-Native System for Autonomous Post-Training Research Develops an open agent environment for evaluating how autonomously language agents can conduct post-training machine learning experiments. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Studies autonomous post-training capability rather than safety or guardrails.
Bellman-Consistent Trajectory RL for LLM Arithmetic Reasoning Applies trajectory Bellman residual minimization to turn sparse rewards into dense token signals for arithmetic reasoning. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Focuses on RL algorithms for arithmetic reasoning, not autonomous multi-step agents.
Bellman-Consistent Trajectory RL for LLM Arithmetic Reasoning Applies trajectory Bellman residual minimization to turn sparse rewards into dense token signals for arithmetic reasoning. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Proposes an RL training method rather than an evaluation framework or benchmark.
Bellman-Consistent Trajectory RL for LLM Arithmetic Reasoning Applies trajectory Bellman residual minimization to turn sparse rewards into dense token signals for arithmetic reasoning. Preprint	Small and efficient models via arxiv · 17 Oct, 06:17	Not relevant Focuses on training RL algorithms rather than model compression or efficient serving.
Bellman-Consistent Trajectory RL for LLM Arithmetic Reasoning Applies trajectory Bellman residual minimization to turn sparse rewards into dense token signals for arithmetic reasoning. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Studies mathematical reasoning optimization, not safety or guardrails.
Approve What, Exactly? Effect Binding and Trusted Rendering for Agent Step-Up Authorization Analyzes authorization security flaws in agent human-in-the-loop approval mechanisms and proposes a tamper-resistant approval gateway. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Proposes a security protocol rather than a standard LLM evaluation benchmark.
OmicsBench: a long-horizon multi-turn benchmark of AI agents on research-study-scale bioinformatics Presents a multi-turn benchmark measuring language agent capabilities on complex, long-horizon bioinformatics workflows. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Focuses on capability benchmarking rather than safety, red-teaming, or guardrails.
Sci$Π$vot: Benchmarking Principal-Investigator Judgment at Pivotal Moments of Scientific AutoResearch Agents Introduces a benchmark evaluating whether language agents can make correct scientific judgment calls derived from human interventions. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Evaluates scientific agent planning accuracy rather than safety or reliability guardrails.
A Jury, Not an Electorate: Label-Free Reliability Estimation for Ensembles of Language Models Estimates language model ensemble member reliability from unlabeled agreement patterns using an expectation-maximization jury formulation. Preprint	Small and efficient models via arxiv · 17 Oct, 06:17	Not relevant Uses ensembles of existing small models rather than proposing model compression or distillation.
A Jury, Not an Electorate: Label-Free Reliability Estimation for Ensembles of Language Models Estimates language model ensemble member reliability from unlabeled agreement patterns using an expectation-maximization jury formulation. Preprint	Multimodal models via arxiv · 17 Oct, 06:17	Not relevant Analyzes textual QA ensembles rather than multimodal vision-language architectures.
A Jury, Not an Electorate: Label-Free Reliability Estimation for Ensembles of Language Models Estimates language model ensemble member reliability from unlabeled agreement patterns using an expectation-maximization jury formulation. Preprint	LLMs for code via arxiv · 17 Oct, 06:17	Not relevant Studies question-answering ensembles rather than code synthesis or software engineering.
A Jury, Not an Electorate: Label-Free Reliability Estimation for Ensembles of Language Models Estimates language model ensemble member reliability from unlabeled agreement patterns using an expectation-maximization jury formulation. Preprint	Retrieval-augmented generation (RAG) via arxiv · 17 Oct, 06:17	Not relevant Deals with unlabelled consensus among models rather than external retrieval or grounding.
A Jury, Not an Electorate: Label-Free Reliability Estimation for Ensembles of Language Models Estimates language model ensemble member reliability from unlabeled agreement patterns using an expectation-maximization jury formulation. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Focuses on ensemble voting methods rather than multi-step tool-using autonomous agents.
SIEVE: Overcoming Topological Obstruction in Equivariant Self-Supervised Learning Introduces an equivariant self-supervised learning framework that avoids topological collapse in 3D point cloud masked autoencoding. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Introduces a self-supervised learning framework rather than LLM benchmarking or evaluation.
SIEVE: Overcoming Topological Obstruction in Equivariant Self-Supervised Learning Introduces an equivariant self-supervised learning framework that avoids topological collapse in 3D point cloud masked autoencoding. Preprint	Computer vision for industrial applications via arxiv · 17 Oct, 06:17	Not relevant Deals with 3D point cloud theory rather than industrial defect detection or visual quality control.
SIEVE: Overcoming Topological Obstruction in Equivariant Self-Supervised Learning Introduces an equivariant self-supervised learning framework that avoids topological collapse in 3D point cloud masked autoencoding. Preprint	AI agents via arxiv · 17 Oct, 06:17	Not relevant Focuses on 3D point cloud representations rather than autonomous planning or tool agents.
SIEVE: Overcoming Topological Obstruction in Equivariant Self-Supervised Learning Introduces an equivariant self-supervised learning framework that avoids topological collapse in 3D point cloud masked autoencoding. Preprint	Small and efficient models via arxiv · 17 Oct, 06:17	Not relevant Addresses representation geometry rather than small models, quantization, or inference acceleration.
SIEVE: Overcoming Topological Obstruction in Equivariant Self-Supervised Learning Introduces an equivariant self-supervised learning framework that avoids topological collapse in 3D point cloud masked autoencoding. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Focuses on self-supervised representations rather than deployment safety, guardrails, or alignment.
SIEVE: Overcoming Topological Obstruction in Equivariant Self-Supervised Learning Introduces an equivariant self-supervised learning framework that avoids topological collapse in 3D point cloud masked autoencoding. Preprint	Multimodal models via arxiv · 17 Oct, 06:17	Not relevant Focuses on geometric 3D point clouds rather than multimodal vision-language or document models.
Probe-and-Resume: When and How to Evaluate Agent Execution Evaluates test-time search in computer-use agents, finding critics perform better when invoked only at predicted final states. Preprint	Small and efficient models via arxiv · 17 Oct, 06:17	Not relevant Focuses on search policies rather than model compression or inference efficiency.
Probe-and-Resume: When and How to Evaluate Agent Execution Evaluates test-time search in computer-use agents, finding critics perform better when invoked only at predicted final states. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Studies agent search policies rather than standardized benchmarks or LLM evaluation suites.
MemComposer: Learning to Assemble Memories for LLM Agents via Reinforcement Learning Uses reinforcement learning to train an agent module that selects, sizes, and orders retrieved memories for downstream tasks. Preprint	Retrieval-augmented generation (RAG) via arxiv · 17 Oct, 06:17	Not relevant Optimizes agent episodic memory assembly rather than general retrieval-augmented document grounding.
MemComposer: Learning to Assemble Memories for LLM Agents via Reinforcement Learning Uses reinforcement learning to train an agent module that selects, sizes, and orders retrieved memories for downstream tasks. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Optimizes memory context assembly rather than AI deployment guardrails, jailbreaks, or safety.
MemComposer: Learning to Assemble Memories for LLM Agents via Reinforcement Learning Uses reinforcement learning to train an agent module that selects, sizes, and orders retrieved memories for downstream tasks. Preprint	Multimodal models via arxiv · 17 Oct, 06:17	Not relevant Deals with agent trajectory memory rather than multimodal image, video, or audio tasks.
MemComposer: Learning to Assemble Memories for LLM Agents via Reinforcement Learning Uses reinforcement learning to train an agent module that selects, sizes, and orders retrieved memories for downstream tasks. Preprint	Small and efficient models via arxiv · 17 Oct, 06:17	Not relevant Focuses on memory assembly policies rather than model quantization, pruning, or serving.
MemComposer: Learning to Assemble Memories for LLM Agents via Reinforcement Learning Uses reinforcement learning to train an agent module that selects, sizes, and orders retrieved memories for downstream tasks. Preprint	LLM evaluation and benchmarks via arxiv · 17 Oct, 06:17	Not relevant Evaluates agent memory assembly rather than establishing general LLM evaluation benchmarks.
MemComposer: Learning to Assemble Memories for LLM Agents via Reinforcement Learning Uses reinforcement learning to train an agent module that selects, sizes, and orders retrieved memories for downstream tasks. Preprint	LLMs for code via arxiv · 17 Oct, 06:17	Not relevant Focuses on agent memory assembly rather than code generation or program repair.
ToolBottleneckBench: Measuring Tool-Agent Performance under Controlled Pressures and Fixed Harness Budgets Measures tool-agent performance failures under controlled harness limits such as completion caps and tool menu visibility. Preprint	Small and efficient models via arxiv · 17 Oct, 06:17	Not relevant Focuses on tool calling and harness limits rather than model compression or quantization.
ToolBottleneckBench: Measuring Tool-Agent Performance under Controlled Pressures and Fixed Harness Budgets Measures tool-agent performance failures under controlled harness limits such as completion caps and tool menu visibility. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Analyzes harness budget limits rather than safety guardrails, jailbreaks, or red-teaming.
ToolBottleneckBench: Measuring Tool-Agent Performance under Controlled Pressures and Fixed Harness Budgets Measures tool-agent performance failures under controlled harness limits such as completion caps and tool menu visibility. Preprint	Retrieval-augmented generation (RAG) via arxiv · 17 Oct, 06:17	Not relevant Focuses on tool-calling harnesses rather than external knowledge retrieval or grounding.
ToolBottleneckBench: Measuring Tool-Agent Performance under Controlled Pressures and Fixed Harness Budgets Measures tool-agent performance failures under controlled harness limits such as completion caps and tool menu visibility. Preprint	LLMs for code via arxiv · 17 Oct, 06:17	Not relevant Studies tool-agent harnesses on enterprise tasks rather than software development or coding.
ToolBottleneckBench: Measuring Tool-Agent Performance under Controlled Pressures and Fixed Harness Budgets Measures tool-agent performance failures under controlled harness limits such as completion caps and tool menu visibility. Preprint	Multimodal models via arxiv · 17 Oct, 06:17	Not relevant Assesses text-based tool execution rather than multimodal vision or document architectures.
Planning Commitment of LLM under Partial Observability Analyzes trade-offs between aggressive and conservative planning horizons in language models operating under partial observability. Preprint	LLMs for code via arxiv · 17 Oct, 06:17	Not relevant Analyzes planning algorithms and PDDL generation rather than software coding or repair.
Planning Commitment of LLM under Partial Observability Analyzes trade-offs between aggressive and conservative planning horizons in language models operating under partial observability. Preprint	Retrieval-augmented generation (RAG) via arxiv · 17 Oct, 06:17	Not relevant Studies planning commitment in environments rather than retrieval-augmented document grounding.
Planning Commitment of LLM under Partial Observability Analyzes trade-offs between aggressive and conservative planning horizons in language models operating under partial observability. Preprint	AI safety and reliability in deployment via arxiv · 17 Oct, 06:17	Not relevant Studies agent planning efficiency trade-offs rather than AI safety, guardrails, or jailbreaks.
Planning Commitment of LLM under Partial Observability Analyzes trade-offs between aggressive and conservative planning horizons in language models operating under partial observability. Preprint	Multimodal models via arxiv · 17 Oct, 06:17	Not relevant Investigates decision-making policies rather than multimodal vision-language architectures.
Planning Commitment of LLM under Partial Observability Analyzes trade-offs between aggressive and conservative planning horizons in language models operating under partial observability. Preprint	Small and efficient models via arxiv · 17 Oct, 06:17	Not relevant Examines planning commitment trade-offs rather than model compression or efficient serving.

5,041–5,100 of 17,313← Previous Next →