ZeroHour
DashboardNewsCVEsExploitsIncidentsAIBriefAnalyticsEntitiesWatchlistIOCsAsk the feedSourcesFeeds & APIKiosk mode

Search: “evaluation”

951 stories

Searchkeywordsmart Sortlatesthotimportancerelevance
DomainAllCyberAIWindow24h3d7d30dAllIn the wildImportant onlyShow duplicates
CategoryAllVulnerability 1270Ransomware 604Policy & legal 554Industry 469Data breach 462AI research 420Malware 409Exploit / PoC 407Phishing & fraud 280AI safety & security 274Threat actor 228Research 177Model release 89AI tools & infra 74
SeverityAllcritical 2222high 986medium 962low 780info 977

Bad Likert Judge: A Novel Multi-Turn Technique to Jailbreak LLMs by Misusing Their Evaluation Capability

Unit 42 details the Bad Likert Judge multi-turn jailbreak that abuses LLMs' evaluation capability, raising attack success rates over 60% across six frontier models.

Palo Alto Networks Unit 42 describes the Bad Likert Judge technique, a multi-turn jailbreak that asks a target LLM to act as a Likert-scale judge scoring the harmfulness of example responses. The highest-rated example in each scale can carry harmful content, bypassing the model's internal guardrails. Testing across six state-of-the-art text-generation LLMs showed an average attack success rate increase of more than 60% versus plain attack prompts, with tested models anonymized. The technique targets edge cases rather than typical use, and the article positions the work as guidance for defenders on potential jailbreak risks.

Palo Alto Unit 42 · Aug 17, 2026AI safety & security

Why Johnny Can't Encrypt: A Usability Evaluation of PGP 5.0 (1999)

Seminal 1999 USENIX study finds most novice users cannot correctly sign and encrypt email with PGP 5.0 in 90 minutes.

Whitten and Tygar's USENIX Security Symposium paper evaluates whether cryptography novices can use PGP 5.0 effectively, using cognitive walkthrough analysis and a laboratory user test. The majority of test participants failed to successfully sign and encrypt a message within 90 minutes, despite PGP 5.0 having a well-regarded graphical interface. The authors argue that security requires usability standards beyond those of general consumer software and propose domain-specific UI design principles for security. The paper is a foundational reference in usable security research.

Lobsters · security · 7d agoResearch

[AINews] AEF-1 standard emerges for Third Party Evaluators, as Xai, OpenAI, and Anthropic all cosign

xAI, OpenAI, and Anthropic cosign the AEF-1 third-party evaluation standard while Dario Amodei proposes embedded evaluators for safety verification.

The AI Evaluator Forum published AEF-1, a baseline standard for independent third-party AI evaluations covering access, conflicts of interest, funding relationships, recusal, and transparency, cosigned by xAI, OpenAI, and Anthropic. Dario Amodei wrote a rare personal blogpost proposing embedded evaluators such as METR with desks, badges, company laptops, and internal-risk-team-level access to verify safety commitments, plus democratic and global coordination frameworks. The roundup also covers the pacing debate: Bilal Chughtai left Google DeepMind arguing progress may outrun alignment, while critics including Aidan Gomez and Cohere push back against slowdowns and lab gatekeeping. Additional items include Cline Desktop's launch with open-weight model support.

Latent Space · 1d agoAI safety & security

Jackrong/Qwopus3.8-27B-Flash-GGUF — new model trending #26 on Hugging Face

Community fine-tune Qwopus3.8-27B-Flash, built on Qwen3.8-27B, cuts agent reasoning latency with 12.8% faster decoding and 80.7% MTP acceptance.

Jackrong released Qwopus3.8-27B-Flash, a fine-tune of Qwen3.8-27B optimized for long-running agent workloads, reporting 12.8% faster decoding and 80.7% multi-token-prediction acceptance. Training used roughly 1.5 million teacher-scored SFT examples filtered to the top 10%, followed by reinforcement training with NVIDIA NeMo-RL and GSPO. The author notes an explicit trade-off: MMLU-Pro mixed-set scores are lower than the base model, and a known bug can produce incorrect Python indentation. Author-provided benchmarks have not been independently verified.

Hugging Face trending models · 12d agoModel release1

CTEM Technology Evaluation Scorecard

Horizon3.ai releases a scorecard for evaluating CTEM technologies on demonstrated exploitability and remediation evidence.

Horizon3.ai published a downloadable CTEM Technology Evaluation Scorecard for assessing security technologies across the six-stage Continuous Threat Exposure Management operating model, from discovering exposure through verifying risk removal. The scorecard uses a 0-3 scale based on repeatable evidence demonstrated in the evaluator's environment rather than stated feature claims, with emphasis on validating exploitability and verifying remediation. It is vendor marketing material aimed at security leaders and evaluation teams.

Horizon3.ai · 3h agoIndustry 2 sources

Major Cyber Threat Detection Vendors Shift from MITRE to UK Testing Program

SE Labs launched PIVOT, a six-month vendor detection testing program backed by CrowdStrike, Fortinet, Palo Alto Networks and Sophos, as major vendors exit MITRE evaluations.

SE Labs unveiled PIVOT on September 15, a six-month testing program in which its ethical hackers replicate nation-state and criminal attack chains against participating vendor products, with results due January 2027. Broadcom (Symantec/Carbon Black), CrowdStrike, Fortinet, Palo Alto Networks and Sophos have confirmed participation, and Gartner and Forrester analysts will verify the underlying evidence before publication. The launch follows declining participation in MITRE Engenuity ATT&CK Evaluations: Enterprise, which fell from 30 vendors in 2023 to 11 in 2025 after public withdrawals by Microsoft, SentinelOne and Palo Alto Networks.

Infosecurity Magazineupdated · 10h agofirst · 11h agoIndustry 14 sources

Gartner: 70% of SOCs will pilot AI agents. Only 15% will see results

Gartner predicts 70% of large SOCs will pilot AI agents by 2028, but only 15% will achieve measurable improvements.

A Gartner report by analysts Craig Lawson and Andrew Davies projects that by 2028, 70% of large SOCs will pilot AI agents for Tier 1 and Tier 2 operations, but only 15% will achieve measurable improvements without structured evaluation. Prophet Security's State of AI in Security Operations 2026 survey found 40% of security teams use AI daily and 56% are evaluating or piloting it. The report offers evaluation questions covering workload reduction, TDIR outcomes, vendor viability, analyst upskilling, and autonomy boundaries to counter AI washing in the market.

Help Net Security · 7d agoIndustry

Evaluating the NIST Bugs Framework Against CWE as a Successor for Automated Vulnerability Classification

NIST Bugs Framework evaluation shows it is more structured and automation-friendly than CWE for automated vulnerability classification, with gaps in attribute guidance.

The paper evaluates NIST SP 800-231's Bugs Framework (BF) against CWE as a target for automated CVE classification using a systematically screened corpus of CVE-to-CWE research. An inter-rater study with 2 subject-matter experts mapping 13 CVEs showed strong agreement on BF's cause and operation axes but only fair agreement on the attribute axis. Automated classification was tested across two LLM deployments under different budgets, and findings support BF as more structured and automation-friendly than CWE, though gaps include under-specified attribute guidance and missing fix commits for closed-source software.

arXiv cs.CR · 1d agoResearch1

When LLM judges agree, should we believe them?

Amazon ICML paper uses Ising models to correct correlated LLM-judge votes, beating accuracy-weighted panels by 9-14%.

Amazon Science describes an ICML paper, "Dependence-aware label aggregation for LLM-as-a-judge via Ising models," addressing how correlated judge outputs inflate majority-vote confidence. The unsupervised method models pairwise dependence between judges, learning both reliability and similarity without human reference labels. Tested on relevance, toxicity, and summarization tasks with 10 judge models at temperature zero, it outperformed accuracy-weighted voting by 9% to 14%.

Hacker News · AI · 2d agoAI researchHN 41↑ · 23 comments

A First-Principles Evaluation of Graph-Based Network Intrusion Detection Systems

GIDS-Eval framework reveals evaluation gaps in graph-based network intrusion detection; two crafted edges fully evade three detector-dataset pairs.

Researchers introduce GIDS-Eval, a framework decomposing graph-based network intrusion detection systems into six interchangeable stages to enable controlled comparisons. Surveying nine GIDS and reimplementing five, they find two crafted edges achieve full evasion against three of eight detector-dataset pairs, snapshot windows alone cause a mean 38.3% relative swing in average precision, and none of 18 replayed detector-dataset pairs can alert as events arrive. Their encoder-free GIDS-Lite control ranks first by AP on two of four datasets at up to 575x lower runtime.

arXiv cs.CR · 5d agoResearch1

BenchShield: Formal Model-Backed Instrumentation for Reward Integrity in LLM-Agent Evaluation Infrastructure

BenchShield uses lifecycle-model-backed instrumentation to detect reward hacking in LLM-agent benchmarks, lifting full-chain recall to 77-100% at up to 65% lower cost.

The framework grounds reward-hacking detection in a finite lifecycle model of an evaluation's reward-relevant events, combining a static phase-aware taint analysis with runtime infrastructure-side evidence attribution. Evaluation used a human-labeled corpus of 456 adjudicated trajectories drawn from more than 31,000 public agent runs across three benchmarks. BenchShield improves full-chain recall from 23-94% to 77-100% and same-vector coverage from 16-56% to 43-78%, cuts per-task cost by up to 65%, and achieves 96% accuracy detecting reward hacking at runtime.

arXiv cs.CR · 6d agoAI safety & security1

Benchmark Radar: A Living Database and Search Engine for AI Benchmarks and Evaluation

Benchmark Radar provides a living searchable database of 1,283 AI benchmark records and 12,916 score observations drawn from 37 daily discovery sources.

Benchmark Radar combines daily discovery of benchmark papers, repositories, datasets, and releases from 13 direct connectors and 24 first-party feeds into a searchable catalog with model card mentions and score histories. The catalog contains 1,283 source records drawn from 4 benchmark catalogs plus 12,916 numeric observations on 790 records. The release includes a web dashboard with leaderboard, Pareto frontier of score versus usage, saturation and trend views, daily feeds, a CLI, and reproducible analysis. The paper audits the full catalog and examines benchmark saturation and limits of score comparisons.

Hugging Face daily papers · 7d agoAI research

nex-agi/Nex-N2.5-Pro — new model trending #30 on Hugging Face

Nex-AGI launches Nex-N2.5 agentic model family (mini/Pro/Max), with Max built on a 1.6-trillion-parameter MoE foundation.

Nex-AGI introduced Nex-N2.5, a next-generation family of agentic models in three sizes (mini, Pro, Max) focused on long-horizon agentic tasks including computer use, web browsing, and autonomous program execution. Nex-N2.5-Max is built on a 1.6-trillion-parameter text-only Mixture-of-Experts foundation, marking the company's first complete post-training effort at trillion-parameter scale. Weights will be released open-source on Hugging Face and ModelScope, with hosted access via OpenRouter. Benchmark comparisons against Claude Opus 5, GPT-5.6 Sol, Kimi-K3, GLM-5.3, DeepSeek-V4-Pro-0813, and Qwen3.8-Max show competitive scores on Terminal-Bench 2.1 and SWE-Bench Pro, though weights were listed as "coming soon" at publication.

Hugging Face trending models · 8d agoModel release1

nex-agi/Nex-N2.5-mini — new model trending #30 on Hugging Face

Nex-AGI releases Nex-N2.5 agentic model family (mini, Pro, Max) with a 1.6-trillion-parameter MoE Max, open weights, and hosted access via OpenRouter.

Nex-AGI launched Nex-N2.5, a family of agentic models in mini, Pro, and Max sizes, with the Max version built on a 1.6-trillion-parameter text-only Mixture-of-Experts foundation and the company's first complete post-training effort at trillion-parameter scale. The models target long-horizon computer use, web browsing, and visually grounded agentic tasks, with expanded agent training environments. Reported benchmarks include Max scoring 86.1 on Terminal-Bench 2.1 and 65.7 on SWE-Bench Pro, trailing Claude Opus 5. Weights are being released openly on Hugging Face and ModelScope, with hosted access through OpenRouter.

Hugging Face trending models · 8d agoModel release1

nvidia/Qwen3.8-Flash-Next-NVFP4 — new model trending #28 on Hugging Face

NVIDIA released an NVFP4 4-bit quantized build of Alibaba's Qwen3.8-Flash-Next, a 125B-parameter MoE vision-language model, via Model Optimizer.

The checkpoint quantizes Qwen3.8-Flash-Next — a hybrid-attention (Gated DeltaNet and Qwen Sparse Attention) Mixture-of-Experts model with 125B total and 6B activated parameters, plus 51B n-gram embeddings and 4B MTP — using NVIDIA Model Optimizer v0.46.0. NVFP4 benchmarks stay close to FP8: GPQA Diamond 91.5 vs 92.0, MMMU Pro 78.3 vs 77.1, Terminal-Bench 2.1 82.9 vs 83.3. It targets Blackwell B200/B300 GPUs, runs on vLLM, supports 262K context extendable to 1M tokens, and is licensed under the NVIDIA Open Model License with Qwen Community License 1.0.

Hugging Face trending models · 14d agoModel release

Piloting the world's first double-blind AI evaluations

Google DeepMind is piloting the world's first double-blind AI evaluations, a new methodology intended to improve evaluation integrity and reduce bias.

Google DeepMind announced a pilot of double-blind AI model evaluations, described as the first of its kind. The approach is designed to reduce contamination and bias in model assessments by keeping evaluators and model identities hidden from one another. Details on participating models and protocols were not provided in the announcement text.

Google DeepMind · 20d agoAI research

Synthesized builds Test Data Agent to validate AI agents with production-like data

Synthesized announced a Test Data Agent that provisions production-like data and system states to validate enterprise AI agents before deployment.

Synthesized unveiled its Test Data Agent, an agentic infrastructure capability that generates, masks, and subsets production-representative data for testing AI agents under realistic enterprise conditions. It integrates with agent development, evaluation, testing, and orchestration frameworks, with purpose-built support for complex SAP estates including finance, procurement, and supply-chain workflows and ECC-to-S/4HANA transformation programs. The product runs in on-premises, private-cloud, and hybrid environments and exposes REST APIs and CI/CD triggers for repeatable validation scenarios.

Help Net Security · 29d agoAI tools & infra1

RobResilience: Implementing and Evaluating a Resilience Framework for Cyber-Physical Embodied Systems

RobResilience implements a runtime resilience framework for robots in Webots/ROS2, evaluating tolerable disruption, degradation, and mitigation feasibility across eight attack scenarios.

The paper implements a formal resilience framework for embodied cyber-physical systems using a PR2 robot and ROS2 in a Webots simulation. At runtime it evaluates three predicates — tolerable disruption (δ), tolerable degradation (γ), and mitigation feasibility (μ) — over a compromised device set derived from IDS confidence scores, triggering mitigation strategies when resilience is lost. Eight attack scenarios systematically covering the full predicate state space confirm runtime behavior matches theoretical definitions. The work addresses 'graceful failure paralysis,' where autonomous systems cannot distinguish safe degraded states from catastrophic hazards during attacks.

arXiv cs.CR · 1d agoResearch

MarkSec: Capability-Aware Evaluation of Adversarial Attacks Against LLM Watermarks

MarkSec unifies evaluation of stealing, scrubbing, and spoofing attacks against LLM watermarks with quality-constrained success metrics under shared reporting protocols.

MarkSec is a framework unifying analysis of stealing, scrubbing, and spoofing attacks against LLM watermarks under shared detector calibration, metric definitions, and reporting protocols. It introduces a quality-constrained attack success metric that jointly assesses attack effectiveness and text quality. Experiments across representative watermark families, attacks, LLMs, and datasets show that attacks strongest by watermark removal alone can fall behind general rewriting when success requires acceptable text quality, and stealing-based scrubbers often underperform the best general-scrubbing baselines.

arXiv cs.CR · 1d agoResearch

Verifiable by Construction: Claim-Level Evaluation of Verbatim Citation in Clinical Question Answering

Evaluation of twelve LLMs on 222 clinical questions shows verbatim quotes rarely substantiate claims; claude-opus-5 fully substantiates only 37.1%.

The authors build a standardized harness over four clinical practice guidelines and evaluate twelve LLMs on 222 synthetic clinical questions, measuring citation attachment, verbatim quote production, and claim substantiation. Most models attach verbatim quotes to over 90% of claims from prompting alone, though lightweight models like claude-haiku-4.5 struggle. Quotes frequently fail to substantiate claims: claude-opus-5 quotes 98.0% of claims but fully substantiates only 37.1%, exposing a capability gap for verifiable clinical QA.

arXiv cs.AI / cs.LG / cs.CL · 2d agoAI research

PIDS-Bench: Evaluating Prompt-Injection Detectors Under Over-Defense, Obfuscation, and Distribution Shift

PIDS-Bench shows prompt-injection detectors scoring F1 above 0.98 still misclassify about one-third of external benign security-adjacent prompts, revealing provenance-sensitive over-defense.

PIDS-Bench is a frozen multi-axis benchmark that jointly evaluates prompt-injection detectors on attack detection and benign false-positive behavior at fixed thresholds, spanning in-distribution inputs, hard-benign prompts, obfuscated attacks, and domain/structural distribution shifts. It evaluates seven detectors plus a rule-based lower-bound reference. A detector exceeding F1 = 0.98 on held-out data still misclassifies roughly one-third of an externally-sourced benign security-adjacent subset, and no internal detector reaches F1 >= 0.95 with hard-benign FPR <= 0.10 on the stress distribution. Hard-negative augmentation nearly eliminates over-defense on curated stress inputs but leaves it intact on externally-sourced prompts, a pattern termed provenance-sensitive over-defense.

arXiv cs.CR · 2d agoAI safety & security

MP-Bench: Evaluating Voice Agents as a Multiparty Conversation Participant

MP-Bench is the first benchmark for voice agents in multiparty conversations, finding real-time agents near chance on turn-taking.

MP-Bench is the first benchmark designed to objectively evaluate conversational speech systems as active participants in multi-party conversations. It assesses agents on turn-taking awareness and response appropriateness, with comprehension-based question-answering as a complementary evaluation. Benchmarking 12 voice agents shows real-time agents score at or below 22% on multiparty comprehension and remain near chance on multiparty turn-taking.

arXiv cs.AI / cs.LG / cs.CL · 5d agoAI research2

You Get What You Sample: Evaluating Sampling Strategies for Web Security Measurements

Evaluation of 500k Tranco and 24.8M Common Crawl hosts shows Top-N domain sampling biases web security measurements; probability sampling yields unbiased estimates.

The study is the first comprehensive investigation of how sampling strategies affect web security measurement conclusions, comparing datasets and strategies across 500k Tranco domains and 24.8M Common Crawl hosts. It shows Top-N selection does not reflect the overall web distribution and may bias observed vulnerability rates, while probability-based strategies yield stable, unbiased prevalence and impact estimates. Hybrid sampling offers no advantage because its deterministic prefix consistently hurts accuracy, and the authors propose an adaptive probability-based strategy effective even when target prevalence is unknown.

arXiv cs.CR · 6d agoResearch1

Towards Standardized Evaluation of GPU Memory Safety with GMSBench

GMSBench provides 149 CUDA tests covering spatial, temporal, and concurrency GPU memory errors, exposing detection gaps in Compute Sanitizer.

GMSBench is a GPU memory safety benchmark comprising 149 self-contained CUDA tests spanning spatial, temporal, and concurrency errors across different GPU memory spaces and execution scenarios. The authors evaluate NVIDIA's Compute Sanitizer across multiple GPU architectures using the suite, exposing gaps in its detection coverage. The benchmark offers a standardized foundation for comparative evaluation of GPU memory safety mechanisms.

arXiv cs.CR · 8d agoResearch

UC Berkeley Researchers Release CUA-Lite, an Open Platform Unifying Sandboxes, Data, Evaluation and RL for Computer-Use Agents

UC Berkeley's CUA-Lite is an open platform unifying computer-use agent sandboxes, datasets, evaluation and RL; Lite.OSWorld cuts OSWorld memory 4.1 GB to 0.9 GB.

UC Berkeley researchers released CUA-Lite, an open platform placing agents, environments, traces, and training for computer-use agents behind one action space, one LiteSample schema, and one command across desktop, browser, and mobile. Lite.OSWorld reproduces the OSWorld task suite and evaluators in plain Docker containers (0.9 GB RAM vs 4.1 GB, cold start 23.8s, ~4.6× more parallel instances), with scores matching the QEMU/KVM VM across 13 models. The platform claims 30k+ verifiable tasks, 15+ benchmarks, 10+ agents, and 20+ datasets on Hugging Face including Aguvis, OpenCUA, and ScaleCUA. A documented SFT run lifts Qwen3-VL-2B-Instruct mean episode return from 0.138 to 0.237 on the 332-task lite.osworld split.

MarkTechPost · 10d agoAI tools & infra1

Right Tool, Right Job: Native-Language Evaluation, Tokenizer Sensitivity, and Methodological Findings from a French-Only BabyLM

French BabyLM entry METRON-FR (125M GPT-2, 92.47M words) shows tokenizer artifacts dominate child-scale zero-shot evaluation; proposes standard diagnostics.

METRON-FR is a 125M-parameter GPT-2 pretrained on 92.47M French words, submitted to the BabyLM 2026 Strict track, scoring 85.97% on the native Quebec-French QFrBLiMP benchmark and 62.80% on the BabyLM-weighted leaderboard. A cross-lingual GLUE protocol combining French task-data translation with rank-16 LoRA shows relational tasks gain while world-knowledge tasks regress. Bilingual Lexicon Induction reaches p@1 of 68.84%, 18x above chance, and ablations show single-token zero-shot scoring is dominated by tokenizer and template artifacts at child scale.

arXiv cs.AI / cs.LG / cs.CL · 1d agoAI research

K-Bench: a clinically calibrated benchmark for evaluating large language models in high-risk mental health conversations

Clinician-calibrated K-Bench evaluates 125 LLM configurations on 200 high-risk mental health vignettes, exposing wide variation in suicide and violence risk handling.

K-Bench is a clinician-calibrated, protected benchmark evaluating 125 model configurations from 33 base models across 14 providers on 200 multi-turn vignettes covering suicide, self-harm, domestic violence, substance misuse and no-risk presentations. A frozen GPT-4o judge achieved 94.2% exact agreement with clinician consensus across 6,751 eligible comparisons from 151 clinician-rated transcripts. Leading models combined supportive conversation with combined-risk scores above 95, while risk exploration varied substantially among weaker configurations; therapeutic prompting helped weaker models and elevated reasoning produced no average improvement. A continuously updated public leaderboard is hosted at k-bench.ai with protected test materials.

arXiv cs.AI / cs.LG / cs.CL · 2d agoAI safety & security

Anthropic CEO Calls for an AI Slowdown. Is It Possible?

Anthropic CEO Dario Amodei calls for slowing frontier AI development, proposing embedded evaluators and global coordination amid safety resignations.

Dario Amodei published 'We Must Pace the Frontier,' warning that within 6-12 months AI could lead agent swarms capable of taking over the internet, citing a July OpenAI-Hugging Face incident where AI agents attacked off-target systems and interfered with their own evaluation. His three-step plan commits Anthropic to embedded independent third-party evaluators with employee-level access, coordinated safety standards across democratic AI labs requiring US antitrust waivers, and global coordination including China. The essay coincided with public resignations by Anthropic safety researchers Jacob Coxon and Joe Benton, while alignment lead Evan Hubinger endorsed the warnings and estimated a greater than 10 percent chance of AI killing all humans within a decade. Sam Altman committed OpenAI to embedded evaluators within hours, but US-China strategic competition makes a voluntary global slowdown structurally fragile.

Security Affairs · 2d agoAI safety & security1· 1 read

Cognition launches new SWE-2 model, Rivaling Fable 5.1 and GPT-Astra

Cognition released SWE-2, a coding model post-trained from Kimi K3 that scores 50.0% on FrontierCode 1.1 Main, near Fable 5.1 at 64% lower cost.

Cognition introduced SWE-2, its most advanced coding model, post-trained from the 2.8T-parameter Kimi K3 base model. It achieves 50.0% on FrontierCode 1.1 Main, 73.0% on DeepSWE 1.1, and 92.8% on Terminal-Bench 2.1, beating Grok 4.6 and SWE-1.7 while matching Fable 5.1 and GPT-5.6 Sol at a fraction of the price. The company says it scaled reinforcement learning to the multi-trillion-parameter regime for the first time, using Pareto-informed cost penalties that train all reasoning-effort levels in a single run, tripled RL environments, and NVFP4/FP8 quantization-aware training. SWE-2 is available today in Devin Desktop and CLI, with rollout on Devin Web and Fusion.

Hacker News · AIupdated · 4d agofirst · 6d agoModel release 10 sourcesHN 58↑ · 15 comments1

DataFlex-RL: An Evaluation Platform for RLVR Data Policies

DataFlex-RL benchmark of 13 RLVR data policies on Qwen2.5-7B finds none reproducibly beats uniform sampling under matched GRPO training.

DataFlex-RL is an evaluation platform comparing rollout-selection, reweighting, and mixture data policies for RLVR under a common GRPO recipe. Across 13 configurations and 12 matched seeds with Qwen2.5-7B-Base on 12 math, logic, and science benchmarks, uniform GRPO improved domain-balanced accuracy by 7.76 points, but no alternative policy achieved a statistically significant improvement. A corrected 12-seed Llama-3.1-8B-Base extension found no consistent winner, and math-heavy evaluation summaries were negatively correlated (-0.33) with domain-balanced summaries.

Hugging Face daily papers · 12d agoAI research