ZeroHour
DashboardNewsCVEsExploitsIncidentsAIBriefAnalyticsEntitiesWatchlistIOCsAsk the feedSourcesFeeds & APIKiosk mode

Search: “measurement study”

123 stories

Searchkeywordsmart Sortlatesthotimportancerelevance
DomainAllCyberAIWindow24h3d7d30dAllIn the wildImportant onlyShow duplicates
CategoryAllVulnerability 257Ransomware 231Data breach 201Policy & legal 164Industry 151Malware 140Phishing & fraud 122Exploit / PoC 99Threat actor 85Research 81AI research 45AI safety & security 42AI industry 13Advisory 10
SeverityAllcritical 634high 325medium 242low 339info 126

Approval Integrity and Recovery in LLM Answer Publication

Study measures approval integrity in Lightcap LLM answer publication, finding the 14B response-act checker accepts 291 of 302 unsupported answers.

The study evaluates exact-content binding, authorization freshness, and checkpoint recovery in Lightcap's publication enforcement using 3,600 assessments over 900 human-annotated RAGTruth responses from three Ministral models. The production 14B response-act checker accepts 291 of 302 unsupported answers versus 41 for a direct-grounding baseline, with supported-answer retention of 95.2% versus 66.9%. A stateful recheck-recovery policy increases exact-match error by 9.23 percentage points relative to initial checkpoints, and controlled evidence-fingerprint changes expose asymmetric freshness enforcement between publication and recovery. A separate BIPIA prompt-injection experiment records zero target insertions among 266 valid editor outputs.

arXiv cs.CR · 2d agoAI safety & security

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

When companies get specific about AI, revenue growth looks different

A Carnegie Mellon and Larridin study of 564 companies links specific AI disclosures to 8 points higher revenue growth, not margins.

Researchers at Carnegie Mellon University and Larridin analyzed 564 companies across 12 sectors using 478 10-K filings, more than 30,000 classified job postings and an AI maturity tracker. Companies at the top of 'narrative concreteness', meaning they named deployed AI systems with measurable outcomes, were associated with 8.0 percentage points higher year-over-year revenue growth, while no public AI signal predicted operating margins or risk-adjusted stock returns. AI infrastructure suppliers outperformed matched peers by about 32 percentage points over four months. The authors stress the findings show association, not causation.

Help Net Security · Aug 17, 2026AI industry

Searching for New Physics with Reinforcement Learning

Researchers apply reinforcement learning to identify SMEFT operators explaining particle physics anomalies, reproducing and improving known CDF W-mass results.

The paper introduces a reinforcement learning method to search the large Standard Model Effective Field Theory (SMEFT) operator space for explanations of measurement anomalies. It was validated on the CDF W-mass anomaly, reproducing and improving known results, then applied to a harder multi-anomaly scenario. RL efficiently navigates complex loop-level operator correlations that bias human-driven phenomenological analysis.

arXiv cs.AI / cs.LG / cs.CL · 7d agoAI research

"Shut Up and Let Me Enjoy My Otome": Understanding and Measuring the Toxicity in Otome Game Communities

First large-scale study finds 22.20% of Weibo otome game posts toxic versus 3.71% on Reddit, with LLM detectors reaching 0.82 F1.

Researchers present the first large-scale measurement of toxicity in otome game communities, introducing OtomeSCAN, which collected and analyzed 620,045 posts from Weibo and Reddit over 18 months. They manually annotated 4,308 posts, identified eight target groups, and evaluated seven toxicity detectors, with their best LLM-based model reaching F1-scores of 0.82 on Weibo and 0.78 on Reddit. The study found 22.20% of Weibo posts were toxic versus 3.71% on Reddit, and toxicity rose to 37.09% within 72 hours during an external attack on Weibo. The authors also flagged 191 potential-coordination clusters, 64.40% of which targeted game developers.

arXiv cs.CR · 9d agoResearch

When Agents Slow Down: Understanding LLM Agents' Test-Time Strategies via Elo-per-token Analysis

Elo-per-token analysis shows LLM agents' marginal gains drop below independent sampling at scale; parallel sessions beat one long session.

The paper proposes Elo-per-token analysis, using a Bradley-Terry model to measure how agent performance scales with token budget on open-ended tasks with continuous scoring. Across four agents and four benchmarks with sessions up to 100M tokens, agents initially convert tokens to Elo faster than independent sampling but eventually slow below the linear-in-log-compute reference. The authors define a scaling inflection point and show that splitting 100M tokens across parallel sessions on FrontierCS Polyomino Packing gains +264 Elo over one long session and +355 over ten short sessions. Human contestants on shared AtCoder Heuristic Contest tasks improve superlinearly, indicating headroom over current agents.

Hugging Face daily papers · 3d agoAI research4· 2 reads

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

Certifying Adversarial Robustness of Quantum Classifiers under Known-Readout Query Access

Framework certifies adversarial robustness of quantum classifiers using only measurement statistics and finite-shot outcomes, demonstrated on IBM Quantum hardware.

The paper introduces a measurement-only certification framework for adversarial robustness of quantum classifiers under known-readout query access, requiring no tomography, parameters, or gradients. It returns a lower bound ruling out untargeted errors within a radius and an attack-independent upper bound witnessing an adversarial state, both estimable with finite-sample guarantees. Evaluations show the lower bound tracks exact optima on tractable instances while the upper bound stays informative when standard attacks fail. The method was validated on IBM Quantum hardware using 40 executions of two 8-qubit quantum neural networks.

arXiv cs.CR · 6d agoResearch

Ex-Deepmind VP Vinyals says AI self-improvement is coming but won't trigger an intelligence explosion

Ex-DeepMind research VP Oriol Vinyals says recursive self-improvement is coming but slow, and co-founds Discovery Loop with Jeff Dean to automate research.

Oriol Vinyals, former VP of Research at Google DeepMind who worked on AlphaStar, AlphaCode, and Gemini, argued at Agentic AI Summit 2026 that recursive self-improvement will progress gradually without an intelligence explosion. He identifies idea generation ('research taste') and evaluation as the two biggest bottlenecks, noting benchmarks like SWE-Bench Pro and ML-Bench mostly test the already-working steps and suffer from overfitting and scheming. He is co-founding Discovery Loop with Jeff Dean as CEO, Sanjay Ghemawat, and Quoc Le to automate the full research cycle, starting with AI research.

The Decoder · 5d agoAI industry1

The Price of Sparsity: Sufficient Conditions for Sparse Recovery using Sparse and Sparsified Measurements

Researchers derive sufficient sample-size conditions for recovering sparse binary signals from sparse Gaussian measurements, quantifying an information-theoretic threshold of order slog(p/s)/log(ds/p).

The paper studies support recovery of sparse binary signals from noisy linear measurements. For sparse Gaussian designs, the authors identify sufficient minimal sample sizes for maximum-likelihood recovery in the high-SNR regime d*s/p -> infinity, yielding an information-theoretic threshold of order slog(p/s)/log(ds/p) that makes the price of measurement sparsity explicit. They also show a regime where the sample-complexity loss from sparsity is only logarithmic while computational gains are nearly linear, and prove that for independently sparsified dense Gaussian designs a sample size of order p/ψ² suffices for support recovery at any fixed error level.

Hugging Face daily papers · 9d agoAI research

Evaluating Practical Enumeration and Blocking Attacks on the Snowflake Circumvention System

Ethical measurements enumerated 21,000+ Snowflake proxy IPs across ~1,000 ASes; blocking top 1% of ASes disrupts 30% of Snowflakes.

Researchers tested Snowflake's assumptions that proxy IPs cannot be easily enumerated and that blocking them causes unacceptable collateral damage. Over 48 days of real-world measurements (May-June 2025), malicious-client-style enumeration collected over 21,000 unique proxy IPs across almost 1,000 autonomous systems. Blocking the top 1% of ASes blocks more than 30% of observed Snowflakes while affecting about 2.5% of Tranco Top 1M domains, and the broker's load-aware matching leaks stable high-capacity proxies to attackers. Some proposed mitigations have already been integrated into Snowflake.

arXiv cs.CR · 6d agoResearch1

Do AI Coding Assistants Check Before They Install? A Pre-Registered Demand-Side Audit of Trust Signals in the Research Software Supply Chain

Pre-registered audit finds AI coding assistants verified provenance signals in only 9 of 1,920 trials before installing research software packages.

The study tested whether AI coding assistants check machine-readable trust signals such as SBOMs, signed releases, and provenance attestations before installing six open-source research software projects spanning HPC and quantum computing. Three models under two operating modes produced 1,920 registered trials scored from container logs. Provenance signals were opened in only 9 of 1,920 trials (0.5%) and zero of 384 control trials, with no trial running a verification command. The authors conclude publishing signals is insufficient and verification must be built into the program running the assistant.

arXiv cs.AI / cs.LG / cs.CL · 9d agoAI safety & security

Measuring Language Transfer in Robot Policies: Adding Greek to a Cosmos3 Vision-Language-Action Policy

Researchers added Greek to the Cosmos3 vision-language-action policy using only machine-rephrased instructions, finding bilingual training reaches roughly two fifths of English performance.

The paper studies localizing the open Cosmos3 vision-language-action robot policy to Greek without architectural changes, using machine-rephrased instructions only. Bilingual training yields a consistent 6.7-7.1 point margin over controls on a 90-task, three-seed evaluation suite, while Greek-only training gains at most 2.7 points. Several common evaluation instruments, including color-histogram metrics and single-goal benchmarks, produced false conclusions, and results were dominated by seed variation. The authors recommend building guaranteed-null baselines and replicating low-resource-language results across seeds.

Hugging Face daily papers · 10d agoAI research

Turn it off and on again, but for critical infrastructure

KTH researchers trained a reinforcement-learning intrusion response agent on an emulated segmented OT network that autonomously resets hosts and processes to disrupt intruders.

Researchers at KTH Royal Institute of Technology built a containerized replica of a segmented industrial network, attacked it across 14 days, and captured 40,000 30-second traffic intervals to train a defense agent under partial observability. The agent observes six packet-count numbers per interval, maintains 500 running state hypotheses, and can reset supervisory hosts, water tank processes, or entire subnets, with resets rebooting the target, renewing credentials, and changing its IP. The best agent approached a full-visibility baseline but depends on an assumed attacker behavior model; the testbed comprised three supervisory hosts, two PLCs, two tanks, weak credentials, and CVE-2017-7494 exposure. The team released its implementation and plans validation on a real industrial testbed with a partner.

Help Net Security · 2d agoResearchCVE-2017-74942· 1 read

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

Evaluating Time-Series Foundation Models and Multimodal Dietary Context for CGM Forecasting

Study finds zero-shot time-series foundation models underperform on CGM forecasting; fine-tuned Chronos-Bolt cuts RMSE up to 18.4% and dietary context adds signal.

The paper evaluates time-series foundation models for continuous glucose monitoring forecasting across eight public datasets covering Type 1 diabetes, Type 2 diabetes, and non-diabetes populations. Under a unified protocol, zero-shot foundation models did not consistently outperform baselines like Elastic Net and PatchTST, but lightweight fine-tuning did, with fine-tuned Chronos-Bolt reducing RMSE by 6.5%-18.4% in the T1D cohort and 8.6%-18.2% in the non-diabetes/T2D cohort. A residual-based fusion framework adding dietary context from CGMacros reduced overall RMSE by about 3% and postprandial RMSE by about 15% versus CGM-only baselines.

arXiv cs.AI / cs.LG / cs.CL · 6d agoAI research

Countering misuse of AI: September 2026 / Anthropic

Anthropic publishes threat intelligence on Claude misuse across seven harm areas from December 2025 through August 2026.

Anthropic's Threat Intelligence team details disrupted operations using Claude Haiku, Sonnet, and Opus across cyber operations, influence operations, surveillance, scams, biological misuse, weapons development, and distillation. The report introduces Generative Threat Groups (GTGs), including state-sponsored groups and financially motivated individuals running AI-augmented multi-victim campaigns. It argues AI uplift now collapses the gap between state-sponsored operations and lone actors, aided by frameworks like PentAGI.

Lobsters · securityupdated · 14h agofirst · 5d agoAI safety & security 18 sources1

Vulnerability Localization Benchmark: Measuring Agentic Security Analysis at Repository Scale

VLoc Bench tests 27 language models at locating vulnerable files in 290 repositories; best system reaches 0.229 File F1 and 38.4% of tasks unsolved.

The Vulnerability Localization Benchmark (VLoc Bench) contains 500 real-world vulnerabilities from 290 repositories across six package ecosystems and 147 CWE categories, pairing pre-fix and post-fix repository snapshots. Agents receive only a CWE description and read-only terminal access to identify affected files, and must confirm absence on patched snapshots. The strongest of 27 language models and four static-analysis tools achieves just 0.229 File F1; 38.4% of tasks receive no correct localization, and effective localizers still report unsupported locations on patched repositories.

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

What Does an LLM-Agent Leaderboard Rank Actually Compare?

A methodological study shows close LLM-agent leaderboard rank gaps on SWE-bench and similar benchmarks often do not support superiority claims.

The paper defines an estimand-aware pairwise procedure for comparing agents, checking common support and applying explicit uncertainty rules and practical margins. Across SWE-bench, AgentRewardBench, and tau2-bench, close rank differences are frequently unresolved, and proxy labels or utility rules can change which system is selected. The authors argue a leaderboard score summarizes a released evaluation but does not by itself justify pairwise superiority conclusions.

arXiv cs.AI / cs.LG / cs.CL · 9d agoAI research

Does Your Agent's Memory Survive a Model Upgrade? A Controlled Study of Memory Portability

A controlled study finds agent memory portability varies sharply: fixed-schema knowledge graphs survive model swaps while compressed notes degrade.

The study compares preserving an agent's history as raw long context, RAG chunks, compressed natural-language notes, or fixed-schema knowledge graphs across model upgrades, using 48 synthetic histories and two open-weight sub-10B-parameter models. Fixed-schema KG accuracy changed by only +0.0004 ± 0.0020 after a writer swap, while compressed NOTES shifted asymmetrically by +9.91 or -13.28 percentage points depending on migration direction. Mixed 50/50 embedding migrations captured only 4.96 of an 11.90-point RAG re-embedding gain; 80% of the NOTES deficit came from information lost at construction, and 81% of the RAG deficit from retrieval failures. Store-only repair of NOTES failed to reach 90% recovery in all 48 cases, while retaining raw histories enabled recovery in 34 of 48 for one direction.

arXiv cs.AI / cs.LG / cs.CL · 12d agoAI research1

1Password's AI patching benchmark is misleading

Trail of Bits reanalysis says 1Password's 26% AI clean-fix rate is misleading; 86% of eligible patches blocked exploits.

Trail of Bits critiques 1Password's FLAWED AI patching benchmark, arguing its 26% clean-fix headline mixes trials where agents were instructed to apply wrong fixes (22% of data) with trials that prohibited compiling or testing (36%). Restricting to reasonable conditions, 2,634 of 3,067 patches (86%) blocked the supplied exploit. Trail of Bits also reports 12.5% of 2,265 developer first fixes failed in its own 2024-2026 assessments, and released post-patch-validation and review-walkthrough agent skills.

Lobsters · security · 1d agoResearch

The AI Supply Chain Has a Security Problem, and Much of It Is Sitting on the Open Internet

Researchers counted 36,769 publicly reachable self-hosted AI endpoints, only about 2% behind HTTP authentication, exposing Ollama, vLLM, and Flowise to abuse.

A Mysterium VPN study found 36,769 self-hosted AI endpoints reachable through internet scanning, with only 2.02% returning an HTTP authentication challenge. Open WebUI accounted for 18,529 reachable instances, Ollama for 6,935 fingerprinted hosts, and 5,223 agent-builder and workflow platforms were exposed, often holding API keys, database credentials, and other secrets. The report highlights LLMjacking risk from exposed Ollama APIs, a critical Flowise bug (CVE-2026-40933), leaked n8n tokens, and prior SentinelOne/Censys research finding roughly 175,000 exposed Ollama hosts in 130 countries.

Security Affairs · 5d agoAI safety & securityCVE-2026-409331

338 Million Attack Simulations Reveal The State Of Enterprise Defense

Picus Labs' Blue Report 2026, from 338 million attack simulations, finds defenses strong at the perimeter but blocking only 37% of post-compromise actions.

Picus Labs' fourth annual Blue Report analyzed over 338 million attack simulations from production environments in H1 2026. Average prevention effectiveness rose from 62% to 69%, but only 37% of attacker actions were blocked after compromise, with reconnaissance and credential theft largely missed. IOC-based malware download prevention fell to 50% from 71% in 2024, and Mimikatz credential dumping from LSASS memory was blocked 94% of the time versus 17% from other memory locations and 3% from registry.

Help Net Security · Aug 12, 2026Research

Operational Roles of QRNG-Derived Quantum Entropy in Bitcoin Proof-of-Work Architectures

arXiv study finds quantum-random entropy adds no Bitcoin PoW success advantage but helps assurance in fault and provenance scenarios.

The paper shows replacing classical entropy with QRNG output does not change honest Bitcoin proof-of-work success probability when candidate headers remain distinct. It introduces a reproducible benchmark measuring an entropy-efficiency factor and a reboot-diversity index, finding QRNG value only in assurance-oriented scenarios involving correlated restart faults, namespace reuse, and entropy provenance. Validation is simulation-based, with hardware-in-the-loop testing identified as future work.

arXiv cs.CR · 12d agoResearch

Differentially Private EEG Feature Anonymization: A Privacy-Utility Case Study in Clinical Neurophysiology

Case study applies Gaussian and Laplace differential privacy to clinical EEG features, quantifying privacy-utility trade-offs across three deployment scenarios.

Researchers evaluate subject-level differential privacy for EEG-derived feature representations using Gaussian and Laplace perturbations across client-side, centralized server-side, and decentralized local training scenarios. Utility is assessed with statistical measures and a downstream machine-learning check on clinical neurophysiology data. Results show DP can be integrated into EEG workflows, but mechanism choice, privacy parameters, and sensitivity calibration strongly influence data utility, particularly on small and imbalanced clinical datasets. The study highlights the privacy-utility trade-off in protecting biomedical signals against re-identification and inference risks.

arXiv cs.CR · 6d agoResearch

A Misalignment of AI in Mathematics

25 Fields Medallists including Terence Tao issue a declaration warning that AI companies' benchmark-driven mathematics goals are misaligned with science and society.

Terence Tao announced a declaration signed by 25 initial signatories, all Fields Medallists, warning that AI companies' push to solve mathematical problems as benchmarks is detrimental to the science and misaligned with the mathematical community's goals. The signatories argue that rushed, headline-driven releases of LLM solutions to major problems raise attribution and plagiarism questions and could erode the human process that develops and transmits mathematical ideas. They frame the issue as a broader misalignment between AI outputs and the purpose of intellectual work, affecting other sciences and society at large. The declaration is posted on a public page, invites further signatures in the manner of the Leiden declaration, and has been covered by The Economist.

Hacker News · AIupdated · 5d agofirst · 5d agoAI safety & security 2 sourcesHN 98↑ · 50 comments1

How Does mHC Use Its Residual Streams? Selective Routing and Near-Identity Mixing

Analysis of DeepSeek-V4-Flash shows four-stream mHC residual blocks use only about two streams effectively, with late-layer mixing providing little benefit.

The study examines the four-stream residual pathway of DeepSeek-V4-Flash, finding typical attention or FFN sites effectively use about two streams and that residual mixing is modest, occurring primarily in early layers. Replacing late mixers with identity increases C4 perplexity by only 1.9% while replacing early mixers raises it by 41%. Retaining the three largest routing weights per token increases perplexity by at most 2.7%, showing the model uses only part of the flexibility afforded by the four-stream design.

arXiv cs.AI / cs.LG / cs.CL · 12d agoAI research

IdeaAMBIG: Benchmarking Implementation-Critical Gaps in Research-Idea Specifications

IdeaAMBIG benchmark with 660 instances measures whether LLMs can spot and fix underspecified research-method details for faithful implementation.

Researchers introduce IdeaAMBIG, a benchmark of 660 evidence-grounded instances (163 real-world gaps from reproducibility reports and GitHub issues, 497 controlled synthetic gaps) built from papers, codebases, and reproduction artifacts. It evaluates codification-readiness assessment, defect localization, and clarification action generation. Across 13 LLMs, the best model achieved only 9.6% Macro Defect Recovery Rate on real-world instances but 80.6% clarification success when given the annotated defect. An oracle study showed gold resolutions raise the codification-ready rate from 14% to 98%, identifying defect localization as the main bottleneck.

arXiv cs.AI / cs.LG / cs.CL · 7d agoAI research2

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

Leading mathematicians fear AI is making their field dumber, and warn the rest of us is next

Twenty-five Fields Medal winners, including Terence Tao, warn AI companies' benchmark-driven approach to mathematics is misaligned with the discipline's purpose of understanding.

In a joint statement, 25 Fields Medal winners including Terence Tao, Peter Scholze, and Maryna Viazovska warn that AI systems solving major open problems at machine speed undermine conceptual understanding and raise severe attribution and plagiarism questions. The statement lands amid a controversy accusing OpenAI of training math models to beat researchers to a rumored partial solution of a Millennium Prize Problem. The signatories describe a general threat to intellectual work across professions and call for urgent action by mathematicians, AI companies, and society, while not seeking a ban.

The Decoder · 4d agoAI industry1

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

From Parameters to Answers: How LLMs Retrieve and Use Their Internal Knowledge

Interpretability study traces how Qwen, Llama, and Gemma route query information and internal knowledge across layers when answering questions.

Researchers used layerwise interventions on hidden states to separate query-routing signals from target knowledge in language models, testing Qwen, Llama, and Gemma on country-continent questions with varied answer types. A pair-conditioned request direction strengthens before interventions alter downstream knowledge, opening a causal window while answer-supporting content is still forming. Trajectories differ by model: Gemma shows a partially overlapping mid-layer routing profile, while Llama has no sustained routing-effect window under the same gates.

arXiv cs.AI / cs.LG / cs.CL · 6d agoAI research1

Miles v0.1: Production-Level Post-Training

Radix Ark open-sources Miles v0.1, a full-stack RL post-training framework demonstrated with asynchronous agentic RL on GLM-5.2 744B-A40B across 64 GB300 GPUs.

Miles v0.1 is a full-stack, open-source system for frontier-scale reinforcement-learning post-training, built on slime with rollout engines on SGLang and trainers supporting NVIDIA Megatron-LM and PyTorch FSDP backends plus three weight-synchronization transports. It supports full-parameter RL, LoRA RL, on-policy distillation, supervised fine-tuning, true-on-policy rollout-training alignment, and extends to diffusion models. The end-to-end case study ran fully asynchronous agentic RL on GLM-5.2 744B-A40B for terminal-use coding tasks on 64 NVIDIA GB300 GPUs with a median step time of 263 seconds over the first 30 measured steps. The code is open-sourced on GitHub.

Hugging Face daily papers · 9d agoAI tools & infra

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