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
Project Greenhouse: Progress Toward Fully Open and Sovereign Agentic Search The paper describes training a competitive decoder-only search reranker from scratch using modest computational resources and common datasets. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Presents a trained reranker family rather than an evaluation framework or benchmark study.
Project Greenhouse: Progress Toward Fully Open and Sovereign Agentic Search The paper describes training a competitive decoder-only search reranker from scratch using modest computational resources and common datasets. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Builds a reranker model for search rather than multi-step autonomous planning agents.
Project Greenhouse: Progress Toward Fully Open and Sovereign Agentic Search The paper describes training a competitive decoder-only search reranker from scratch using modest computational resources and common datasets. Preprint	LLMs for code via arxiv · 12 Oct, 16:28	Not relevant Does not address code generation, program repair, or developer tooling.
Project Greenhouse: Progress Toward Fully Open and Sovereign Agentic Search The paper describes training a competitive decoder-only search reranker from scratch using modest computational resources and common datasets. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Focuses on open and sovereign search architectures, not safety, jailbreaks, or guardrails.
When History Helps and Hurts: Selective History Use across Multimodal Turns The paper introduces a 7,000-task benchmark to evaluate how multimodal models selectively use or ignore conversational context. Preprint	Retrieval-augmented generation (RAG) via arxiv · 12 Oct, 16:28	Not relevant Evaluates conversational context management rather than retrieval-augmented generation from external knowledge.
When History Helps and Hurts: Selective History Use across Multimodal Turns The paper introduces a 7,000-task benchmark to evaluate how multimodal models selectively use or ignore conversational context. Preprint	Small and efficient models via arxiv · 12 Oct, 16:28	Not relevant Does not explore model compression, quantization, distillation, or efficient serving.
When History Helps and Hurts: Selective History Use across Multimodal Turns The paper introduces a 7,000-task benchmark to evaluate how multimodal models selectively use or ignore conversational context. Preprint	LLMs for code via arxiv · 12 Oct, 16:28	Not relevant Studies conversational multimodal media understanding rather than software engineering or code.
When History Helps and Hurts: Selective History Use across Multimodal Turns The paper introduces a 7,000-task benchmark to evaluate how multimodal models selectively use or ignore conversational context. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Assesses multimodal conversational history retention rather than autonomous agents with tools.
When History Helps and Hurts: Selective History Use across Multimodal Turns The paper introduces a 7,000-task benchmark to evaluate how multimodal models selectively use or ignore conversational context. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Focuses on conversational benchmarking rather than safety guardrails, jailbreaks, or alignment.
MiMo-V2.6: Scaling Reinforcement Learning Towards Self-Improvement The report presents an omni-modal model family trained by scaling reinforcement learning throughput, environments, and agentic reward compute. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Addresses RL reward hacking internally, but primary contribution is scaling omni-modal RL training.
MiMo-V2.6: Scaling Reinforcement Learning Towards Self-Improvement The report presents an omni-modal model family trained by scaling reinforcement learning throughput, environments, and agentic reward compute. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Introduces a model and RL training methodology rather than a benchmark or evaluation method.
MiMo-V2.6: Scaling Reinforcement Learning Towards Self-Improvement The report presents an omni-modal model family trained by scaling reinforcement learning throughput, environments, and agentic reward compute. Preprint	Retrieval-augmented generation (RAG) via arxiv · 12 Oct, 16:28	Not relevant Focuses on RL training and self-improvement without investigating external retrieval systems.
MiMo-V2.6: Scaling Reinforcement Learning Towards Self-Improvement The report presents an omni-modal model family trained by scaling reinforcement learning throughput, environments, and agentic reward compute. Preprint	LLMs for code via arxiv · 12 Oct, 16:28	Not relevant Mentions coding environments among several domains but focuses fundamentally on foundation model RL training.
MiMo-V2.6: Scaling Reinforcement Learning Towards Self-Improvement The report presents an omni-modal model family trained by scaling reinforcement learning throughput, environments, and agentic reward compute. Preprint	Small and efficient models via arxiv · 12 Oct, 16:28	Not relevant Scales compute up to large context lengths rather than optimizing for small, efficient models.
Specialized Decision Models vs. General-Purpose LLMs: Benchmarking Jev Across Knowledge, Reasoning, and Multilingual Tasks The paper benchmarks a specialized multiple-choice decision model against 19 general-purpose LLMs across knowledge, reasoning, and multilingual tasks. Preprint	Small and efficient models via arxiv · 12 Oct, 16:28	Not relevant Studies decision architectures rather than model quantization, pruning, or efficient inference techniques.
Specialized Decision Models vs. General-Purpose LLMs: Benchmarking Jev Across Knowledge, Reasoning, and Multilingual Tasks The paper benchmarks a specialized multiple-choice decision model against 19 general-purpose LLMs across knowledge, reasoning, and multilingual tasks. Preprint	Multimodal models via arxiv · 12 Oct, 16:28	Not relevant Evaluates text-based multiple-choice benchmarks without visual or multimodal inputs.
Specialized Decision Models vs. General-Purpose LLMs: Benchmarking Jev Across Knowledge, Reasoning, and Multilingual Tasks The paper benchmarks a specialized multiple-choice decision model against 19 general-purpose LLMs across knowledge, reasoning, and multilingual tasks. Preprint	Retrieval-augmented generation (RAG) via arxiv · 12 Oct, 16:28	Not relevant Focuses on multiple-choice decision models without evaluating external document retrieval or RAG mechanisms.
Specialized Decision Models vs. General-Purpose LLMs: Benchmarking Jev Across Knowledge, Reasoning, and Multilingual Tasks The paper benchmarks a specialized multiple-choice decision model against 19 general-purpose LLMs across knowledge, reasoning, and multilingual tasks. Preprint	LLMs for code via arxiv · 12 Oct, 16:28	Not relevant Does not evaluate code generation, software engineering assistants, or program repair.
Specialized Decision Models vs. General-Purpose LLMs: Benchmarking Jev Across Knowledge, Reasoning, and Multilingual Tasks The paper benchmarks a specialized multiple-choice decision model against 19 general-purpose LLMs across knowledge, reasoning, and multilingual tasks. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Benchmarking accuracy and reasoning rather than deployment safety, red-teaming, or guardrails.
Specialized Decision Models vs. General-Purpose LLMs: Benchmarking Jev Across Knowledge, Reasoning, and Multilingual Tasks The paper benchmarks a specialized multiple-choice decision model against 19 general-purpose LLMs across knowledge, reasoning, and multilingual tasks. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Evaluates a non-agentic decision model across benchmarks rather than multi-step tool-using agents.
Natural Language to First-Order Logic LLM-based Autoformalization The paper reviews LLM-based first-order logic autoformalization, distinguishing ontology extraction from logical translation and outlining key benchmarking challenges. Preprint	Retrieval-augmented generation (RAG) via arxiv · 12 Oct, 16:28	Not relevant Focuses on formal logic translation rather than retrieval-augmented generation.
Natural Language to First-Order Logic LLM-based Autoformalization The paper reviews LLM-based first-order logic autoformalization, distinguishing ontology extraction from logical translation and outlining key benchmarking challenges. Preprint	Small and efficient models via arxiv · 12 Oct, 16:28	Not relevant Surveys logic autoformalization rather than model compression or efficient serving.
Natural Language to First-Order Logic LLM-based Autoformalization The paper reviews LLM-based first-order logic autoformalization, distinguishing ontology extraction from logical translation and outlining key benchmarking challenges. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Surveys translation of text to first-order logic rather than multi-step autonomous agents.
Natural Language to First-Order Logic LLM-based Autoformalization The paper reviews LLM-based first-order logic autoformalization, distinguishing ontology extraction from logical translation and outlining key benchmarking challenges. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Surveys logic formalization methods rather than AI safety, jailbreaks, or guardrails.
Natural Language to First-Order Logic LLM-based Autoformalization The paper reviews LLM-based first-order logic autoformalization, distinguishing ontology extraction from logical translation and outlining key benchmarking challenges. Preprint	Multimodal models via arxiv · 12 Oct, 16:28	Not relevant Deals with natural language and formal logic representations without multimodal data.
ILM: An AI-Powered Storytelling Educational Tool The paper introduces an educational tool combining Arabic knowledge graphs and passage retrieval to generate narrative-based questions. Preprint	Multimodal models via arxiv · 12 Oct, 16:28	Not relevant Focuses on text narratives and knowledge graphs without multimodal model components.
ILM: An AI-Powered Storytelling Educational Tool The paper introduces an educational tool combining Arabic knowledge graphs and passage retrieval to generate narrative-based questions. Preprint	LLMs for code via arxiv · 12 Oct, 16:28	Not relevant Applies NLP to educational storytelling rather than code synthesis or software engineering.
ILM: An AI-Powered Storytelling Educational Tool The paper introduces an educational tool combining Arabic knowledge graphs and passage retrieval to generate narrative-based questions. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Presents an educational tool rather than investigating safety, jailbreaks, or robustness.
ILM: An AI-Powered Storytelling Educational Tool The paper introduces an educational tool combining Arabic knowledge graphs and passage retrieval to generate narrative-based questions. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Presents an educational application rather than an autonomous planning agent.
ILM: An AI-Powered Storytelling Educational Tool The paper introduces an educational tool combining Arabic knowledge graphs and passage retrieval to generate narrative-based questions. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Uses LLM-as-a-judge as a minor component rather than studying LLM evaluation methodology.
ILM: An AI-Powered Storytelling Educational Tool The paper introduces an educational tool combining Arabic knowledge graphs and passage retrieval to generate narrative-based questions. Preprint	Small and efficient models via arxiv · 12 Oct, 16:28	Not relevant Presents an educational platform rather than model compression or efficient serving.
A persistent accuracy ceiling in automated verbal deception detection The paper meta-analyzes 25 years of automated verbal deception detection models and identifies a persistent accuracy ceiling around 70 to 75 percent. Preprint	Multimodal models via arxiv · 12 Oct, 16:28	Not relevant Analyzes verbal deception text models without multimodal integration.
A persistent accuracy ceiling in automated verbal deception detection The paper meta-analyzes 25 years of automated verbal deception detection models and identifies a persistent accuracy ceiling around 70 to 75 percent. Preprint	Small and efficient models via arxiv · 12 Oct, 16:28	Not relevant Surveys deception classification models rather than model compression or inference efficiency.
A persistent accuracy ceiling in automated verbal deception detection The paper meta-analyzes 25 years of automated verbal deception detection models and identifies a persistent accuracy ceiling around 70 to 75 percent. Preprint	LLMs for code via arxiv · 12 Oct, 16:28	Not relevant Does not address code generation, program repair, or software engineering.
A persistent accuracy ceiling in automated verbal deception detection The paper meta-analyzes 25 years of automated verbal deception detection models and identifies a persistent accuracy ceiling around 70 to 75 percent. Preprint	Retrieval-augmented generation (RAG) via arxiv · 12 Oct, 16:28	Not relevant Focuses on deception detection rather than retrieval-augmented generation systems.
A persistent accuracy ceiling in automated verbal deception detection The paper meta-analyzes 25 years of automated verbal deception detection models and identifies a persistent accuracy ceiling around 70 to 75 percent. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Meta-analyzes verbal deception detection models rather than agent planning or tool use.
Rehearse Everything, Remember Nothing: Attic-KV Rehearses What Will Be Read The paper shows that whole-context KV cache rehearsal harms compression under tight budgets, proposing targeted rehearsal to preserve answer-relevant entries. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Addresses KV cache compression rather than multi-step agent planning or tool use.
Rehearse Everything, Remember Nothing: Attic-KV Rehearses What Will Be Read The paper shows that whole-context KV cache rehearsal harms compression under tight budgets, proposing targeted rehearsal to preserve answer-relevant entries. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Addresses cache memory efficiency rather than AI safety or guardrails.
Rehearse Everything, Remember Nothing: Attic-KV Rehearses What Will Be Read The paper shows that whole-context KV cache rehearsal harms compression under tight budgets, proposing targeted rehearsal to preserve answer-relevant entries. Preprint	LLMs for code via arxiv · 12 Oct, 16:28	Not relevant Focuses on long-context KV caching rather than code generation or software tasks.
Rehearse Everything, Remember Nothing: Attic-KV Rehearses What Will Be Read The paper shows that whole-context KV cache rehearsal harms compression under tight budgets, proposing targeted rehearsal to preserve answer-relevant entries. Preprint	Multimodal models via arxiv · 12 Oct, 16:28	Not relevant Focuses on text LLM KV cache management without multimodal inputs.
Rehearse Everything, Remember Nothing: Attic-KV Rehearses What Will Be Read The paper shows that whole-context KV cache rehearsal harms compression under tight budgets, proposing targeted rehearsal to preserve answer-relevant entries. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Proposes a KV cache compression algorithm rather than a general evaluation benchmark.
Language-Specific Effects of Tokenizer Choice in Multilingual Language Models The paper trains 123 models across 54 tokenizers to show tokenizer choice matters more for languages with less training data. Preprint	LLMs for code via arxiv · 12 Oct, 16:28	Not relevant Studies multilingual natural language tokenization rather than programming code.
Language-Specific Effects of Tokenizer Choice in Multilingual Language Models The paper trains 123 models across 54 tokenizers to show tokenizer choice matters more for languages with less training data. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Studies multilingual tokenization effects rather than autonomous agent systems.
Language-Specific Effects of Tokenizer Choice in Multilingual Language Models The paper trains 123 models across 54 tokenizers to show tokenizer choice matters more for languages with less training data. Preprint	Retrieval-augmented generation (RAG) via arxiv · 12 Oct, 16:28	Not relevant Analyzes subword tokenizers rather than retrieval-augmented generation systems.
Language-Specific Effects of Tokenizer Choice in Multilingual Language Models The paper trains 123 models across 54 tokenizers to show tokenizer choice matters more for languages with less training data. Preprint	Small and efficient models via arxiv · 12 Oct, 16:28	Not relevant Focuses on multilingual representation quality rather than model compression or efficient serving.
Language-Specific Effects of Tokenizer Choice in Multilingual Language Models The paper trains 123 models across 54 tokenizers to show tokenizer choice matters more for languages with less training data. Preprint	Multimodal models via arxiv · 12 Oct, 16:28	Not relevant Analyzes text tokenizers across languages without multimodal data.
Language-Specific Effects of Tokenizer Choice in Multilingual Language Models The paper trains 123 models across 54 tokenizers to show tokenizer choice matters more for languages with less training data. Preprint	AI safety and reliability in deployment via arxiv · 12 Oct, 16:28	Not relevant Evaluates language modeling performance rather than AI safety or reliability.
When KL Regularization Misfires in Group Policy Optimization Analyzes interaction failures between KL regularization and rewards in group policy optimization and proposes a calibrated surrogate objective. Preprint	Multimodal models via arxiv · 12 Oct, 16:28	Not relevant Deals with language model policy optimization without multimodal inputs.
When KL Regularization Misfires in Group Policy Optimization Analyzes interaction failures between KL regularization and rewards in group policy optimization and proposes a calibrated surrogate objective. Preprint	Small and efficient models via arxiv · 12 Oct, 16:28	Not relevant Focuses on RL loss formulations rather than model compression or inference efficiency.
When KL Regularization Misfires in Group Policy Optimization Analyzes interaction failures between KL regularization and rewards in group policy optimization and proposes a calibrated surrogate objective. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Focuses on reinforcement learning policy optimization rather than multi-step agent architectures.
When KL Regularization Misfires in Group Policy Optimization Analyzes interaction failures between KL regularization and rewards in group policy optimization and proposes a calibrated surrogate objective. Preprint	LLMs for code via arxiv · 12 Oct, 16:28	Not relevant Evaluates mathematical reasoning rather than code generation or program repair.
When KL Regularization Misfires in Group Policy Optimization Analyzes interaction failures between KL regularization and rewards in group policy optimization and proposes a calibrated surrogate objective. Preprint	Retrieval-augmented generation (RAG) via arxiv · 12 Oct, 16:28	Not relevant Addresses RL policy optimization rather than retrieval-augmented generation.
When KL Regularization Misfires in Group Policy Optimization Analyzes interaction failures between KL regularization and rewards in group policy optimization and proposes a calibrated surrogate objective. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Introduces an optimization algorithm rather than an evaluation benchmark or dataset.
SteerablePlex: Can We Steer Full-Duplex Models? Introduces an instruction-following benchmark and training recipe to improve goal adherence in full-duplex conversational models. Preprint	Small and efficient models via arxiv · 12 Oct, 16:28	Not relevant Focuses on steerability and RL training rather than model compression or efficient serving.
SteerablePlex: Can We Steer Full-Duplex Models? Introduces an instruction-following benchmark and training recipe to improve goal adherence in full-duplex conversational models. Preprint	LLMs for code via arxiv · 12 Oct, 16:28	Not relevant Deals with speech dialog control rather than code generation or software engineering.
SteerablePlex: Can We Steer Full-Duplex Models? Introduces an instruction-following benchmark and training recipe to improve goal adherence in full-duplex conversational models. Preprint	Retrieval-augmented generation (RAG) via arxiv · 12 Oct, 16:28	Not relevant Addresses conversational steerability rather than retrieval-augmented generation.
SciTBERT: A family of chronologically consistent language models for scientific and technological language processing Releases BERT models pre-trained with yearly chronological cutoffs to reduce lookahead bias in scientific and patent text mining. Preprint	LLM evaluation and benchmarks via arxiv · 12 Oct, 16:28	Not relevant Develops a domain-specific encoder family rather than a general LLM evaluation benchmark.
SciTBERT: A family of chronologically consistent language models for scientific and technological language processing Releases BERT models pre-trained with yearly chronological cutoffs to reduce lookahead bias in scientific and patent text mining. Preprint	Multimodal models via arxiv · 12 Oct, 16:28	Not relevant Deals with scientific and patent text without multimodal data.
SciTBERT: A family of chronologically consistent language models for scientific and technological language processing Releases BERT models pre-trained with yearly chronological cutoffs to reduce lookahead bias in scientific and patent text mining. Preprint	AI agents via arxiv · 12 Oct, 16:28	Not relevant Presents pretrained transformer encoders rather than multi-step agent systems.
SciTBERT: A family of chronologically consistent language models for scientific and technological language processing Releases BERT models pre-trained with yearly chronological cutoffs to reduce lookahead bias in scientific and patent text mining. Preprint	Retrieval-augmented generation (RAG) via arxiv · 12 Oct, 16:28	Not relevant Introduces scientific encoder models rather than retrieval-augmented generation pipelines.

15,781–15,840 of 17,313← Previous Next →