ZPPO improves distillation to small vision-language models by using binary and negative candidate prompts plus a replay buffer for hard questions, outperforming standard distillation and GRPO on a 31-benchmark suite with largest gains at the 0.8B scale.
arXiv preprint arXiv:2501.17399 , year=
11 Pith papers cite this work. Polarity classification is still indexing.
11
Pith papers citing it
citation-role summary
background 3
citation-polarity summary
roles
background 3polarities
background 3representative citing papers
Joint evaluation of three tool-using LLM agents on 12 realistic tasks finds consistent failures in data awareness, audience awareness, policy compliance, data minimization, and access-boundary awareness, with no agent achieving both correct and safe execution.
FailureScope clusters evaluation probes by cross-model failure patterns via LOMO to produce stable taxonomies that generalize across single-turn, multi-turn, and adversarial regimes, with reported metrics of Kendall's tau 0.81 and AUC 0.88.
Multi-objective genetic prompt optimization creates multi-turn deceptive datasets validated by humans, then detected with 0.89 recall using angular coverage, distance ratio, and linearity features in embeddings.
Self-preference bias persists in rubric-based LLM evaluation even with fully objective, programmatically verifiable rubrics, and can shift subjective medical-chat scores by up to ~10 points.
Evo-Memory is a new streaming benchmark and evaluation framework for self-evolving memory in LLM agents, unifying over ten memory modules and introducing the ReMem pipeline for continual improvement on multi-turn and reasoning datasets.
RaR uses aggregated rubric feedback as rewards in on-policy RL, delivering up to 31% relative gains on HealthBench and 7% on GPQA-Diamond versus direct Likert LLM-as-judge baselines.
MiniMax-M1 is a 456B parameter hybrid-attention MoE model trained with CISPO RL that achieves performance comparable or superior to DeepSeek-R1 and Qwen3-235B on reasoning and software engineering tasks while training in three weeks on 512 GPUs.
ImgEdit supplies 1.2 million curated edit pairs and a three-part benchmark that let a VLM-based model outperform prior open-source editors on adherence, quality, and detail preservation.
LLMs drop 39% in performance during multi-turn conversations due to premature assumptions and inability to recover from early errors.
RAG-DIVE uses an LLM to dynamically generate, validate, and evaluate multi-turn dialogues for assessing RAG system performance in interactive settings.
citing papers explorer
-
Zone of Proximal Policy Optimization: Teacher in Prompts, Not Gradients
ZPPO improves distillation to small vision-language models by using binary and negative candidate prompts plus a replay buffer for hard questions, outperforming standard distillation and GRPO on a 31-benchmark suite with largest gains at the 0.8B scale.
-
An Evaluation of Data Leakage Risks in Tool-Using LLM Agents in Realistic Scenarios
Joint evaluation of three tool-using LLM agents on 12 realistic tasks finds consistent failures in data awareness, audience awareness, policy compliance, data minimization, and access-boundary awareness, with no agent achieving both correct and safe execution.
-
FailureScope: Cross-Regime Behavioral Diagnosis of Language Model Weaknesses
FailureScope clusters evaluation probes by cross-model failure patterns via LOMO to produce stable taxonomies that generalize across single-turn, multi-turn, and adversarial regimes, with reported metrics of Kendall's tau 0.81 and AUC 0.88.
-
Evolving and Detecting Multi-Turn Deception using Geometric Signatures
Multi-objective genetic prompt optimization creates multi-turn deceptive datasets validated by humans, then detected with 0.89 recall using angular coverage, distance ratio, and linearity features in embeddings.
-
Self-Preference Bias in Rubric-Based Evaluation of Large Language Models
Self-preference bias persists in rubric-based LLM evaluation even with fully objective, programmatically verifiable rubrics, and can shift subjective medical-chat scores by up to ~10 points.
-
Evo-Memory: Benchmarking LLM Agent Test-time Learning with Self-Evolving Memory
Evo-Memory is a new streaming benchmark and evaluation framework for self-evolving memory in LLM agents, unifying over ten memory modules and introducing the ReMem pipeline for continual improvement on multi-turn and reasoning datasets.
-
Rubrics as Rewards: Reinforcement Learning Beyond Verifiable Domains
RaR uses aggregated rubric feedback as rewards in on-policy RL, delivering up to 31% relative gains on HealthBench and 7% on GPQA-Diamond versus direct Likert LLM-as-judge baselines.
-
MiniMax-M1: Scaling Test-Time Compute Efficiently with Lightning Attention
MiniMax-M1 is a 456B parameter hybrid-attention MoE model trained with CISPO RL that achieves performance comparable or superior to DeepSeek-R1 and Qwen3-235B on reasoning and software engineering tasks while training in three weeks on 512 GPUs.
-
ImgEdit: A Unified Image Editing Dataset and Benchmark
ImgEdit supplies 1.2 million curated edit pairs and a three-part benchmark that let a VLM-based model outperform prior open-source editors on adherence, quality, and detail preservation.
-
LLMs Get Lost In Multi-Turn Conversation
LLMs drop 39% in performance during multi-turn conversations due to premature assumptions and inability to recover from early errors.
-
RAG-DIVE: A Dynamic Approach for Multi-Turn Dialogue Evaluation in Retrieval-Augmented Generation
RAG-DIVE uses an LLM to dynamically generate, validate, and evaluate multi-turn dialogues for assessing RAG system performance in interactive settings.