Search: “burnout”

50% of CISOs see Mythos as a sign to exit the profession

Survey of 1,001 US and UK CISOs finds 50% consider leaving the profession amid AI-driven pressure, personal liability concerns and burnout.

A survey of 1,001 CISOs in the US and UK found 50% say Anthropic's Mythos and similar cyber-capable AI models have made them consider exiting the profession, while 60% say board pressure to adopt AI is outpacing their governance capability. Average CISO tenure is cited at 18 months, and 78% worry about personal liability for security incidents, up from 56% a year earlier. Executives from BlackBerry, Databricks and IDC discuss D&O insurance, agentic security operations and phased low-risk AI adoption as partial responses.

CSO Online·Sep 9, 08:25 UTC · 7d agoIndustry

We've got one word for it, and it's usually the wrong one

Cisco Talos's Threat Source newsletter critiques 'burnout' terminology, describing four occupational injuries, and flags a UAT-10820 WebDAV stealer campaign at a Ukrainian government organization.

Cisco Talos's Threat Source newsletter argues that 'burnout' is the wrong word for most cybersecurity occupational harm, distinguishing exhaustion, secondary traumatic stress, vicarious trauma, and moral injury based on clinical literature from trauma-exposed professions. The featured disclosure describes a complex WebDAV infection chain found at a Ukrainian government organization, attributed with moderate confidence to the Russian-tracked actor UAT-10820 and assessed as an opportunistic cryptocurrency and credential-stealing operation. The campaign delivers the Amatera stealer alongside ZigCryptoStealer and NetSupport Manager, abusing BNB Smart Chain bulletproof hosting, fake CAPTCHA prompts, a vulnerable driver to kill EDR, and rundll32.exe execution of disguised DLLs with ordinal calls. Weekly headlines also cover a Microsoft Defender 'ShieldCrash' zero-day exploit released after September 2026 Patch Tuesday, a North Korean Linux espionage toolkit backdooring HAProxy, and a multi-hop Google-domain redirect phishing campaign.

Cisco Talos·Sep 10, 18:00 UTC · 6d agoIndustry in the wild1

Quantifying the Engagement Trap: Impact of Short-form Video Recommender Systems on Users with ADHD

A 302-participant study finds engagement-optimized short-form video recommenders cause disproportionate time blindness and distress for users with ADHD.

Researchers ran a stratified Prolific study with 302 participants comparing short-form video recommendation experiences with and without ADHD. Participants with ADHD reported significantly higher time blindness, post-usage regret, and emotional distress despite perceiving recommendations as similarly relevant. The paper proposes neurodiversity-aware, human-centered design interventions to mitigate these algorithmic harms.

arXiv cs.AI / cs.LG / cs.CL·Sep 7, 17:38 UTC · 9d agoAI research

You Can't Prefer Emotions You Don't Sample: Intensity Undershoot in DPO-Tuned LLMs

Study quantifies DPO-tuned LLMs undershooting requested emotional intensity, tracing the gap to candidate-pool extremity rather than conditioning format.

Conditioning an instruction-tuned LLM on continuous valence-arousal targets yields gain of only 0.26 for valence and 0.13 for arousal on Llama-3.1-8B, far below faithful control of 1.0. The authors attribute undershoot to neutral-heavy preference corpora like EmoBank and candidate pools lacking extreme affect, leaving DPO without extreme exemplars. Uniform target coverage with a hotter candidate pool raises valence gain to 0.40 on Llama-3.1-8B and 0.44 on Qwen3-8B, with modest in-distribution cost; arousal gains remain unstable across seeds.

arXiv cs.AI / cs.LG / cs.CL·Sep 7, 17:47 UTC · 9d agoAI research

15 Minutes Saved Per Alert: How a Lean German Manufacturer Protects 10,000 Endpoints with ANY.RUN

A five-person security team at a German manufacturer protecting 10,000 endpoints cut triage time by 15 minutes per alert after adopting ANY.RUN's cloud sandbox.

Philipp Z., Security Lead at a leading German manufacturer, described how a five-person team protects 10,000 endpoints and users using ANY.RUN's Interactive Sandbox in a private cloud. The firm previously relied on a single air-gapped forensic laptop running Flare VM, which caused 5-10 minute setup delays, single-user bottlenecks, and selective triage. The switch reportedly saved roughly 15 minutes per alert and reduced forced wiping and reimaging of user machines. ANY.RUN data cited in the piece puts manufacturing security workloads 22% above other major industries.

ANY.RUN·Sep 9, 07:47 UTC · 7d agoIndustry

Feeling sad about AI

Simon Willison argues developers disheartened by AI coding agents should move past existential crisis and refocus on broader engineering problems.

In a comment on a Hacker News thread titled 'Feeling sad about AI', Simon Willison reflects on the discouragement engineers feel when AI coding agents complete in an hour work that previously took a week. He notes that translating an exact specification into decent code is no longer a unique skill. He counsels that engineers who accept this shift can apply their experience to the larger set of problems they still face.

Simon Willison·Sep 11, 17:28 UTC · 5d agoAI industry2

The exploit gap is closing, and your patch cycle wasn't built for this

Help Net Security·Aug 5, 10:23 UTC · Aug 5, 2026Vulnerability

One resignation turned the embers of AI fear into a wildfire

Interconnects argues a frontier-lab researcher's safety resignation went viral via media coordination, reigniting AI existential-risk discourse.

The essay analyzes why researcher Jacob Coxon's resignation over AI safety risks went viral, aided by a Wall Street Journal exclusive, advocacy-group amplification, and Daniel Kokotajlo's same-day Joe Rogan appearance. It notes Evan Hubinger's >10% extinction-risk figure and criticizes existential-risk discourse for conflating very different meanings of the term. The author assigns near-zero probability to complete extinction but argues concrete risks such as cyberattacks on critical infrastructure and bio-risks merit debate. It also rejects recursive self-improvement forecasts, proposing 'lossy self-improvement' where models excel at math and code but remain limited elsewhere.

Interconnects·Sep 10, 15:28 UTC · 6d agoAI safety & security

Why "Shady AI" is Security's Next Big Governance Problem

Opinion piece argues approved-but-ungoverned AI agent use ('shady AI') is a growing enterprise risk, citing a Meta Sev 1 data-exposure incident.

In March 2026, an approved AI agent at Meta posted a forum answer publicly, exposing sensitive company and user data to unauthorized employees for over two hours and triggering a Sev 1 incident. The author contrasts shadow AI (unapproved tools) with 'shady AI' (approved tools used in unapproved or unanticipated ways), arguing the latter is harder to govern because it operates inside the organization's visibility. A July 2026 SANS survey found 76% of security teams now have a role in governing enterprise AI. The piece recommends governance-by-default environments with built-in permissions, access controls, and monitoring rather than policy and training alone.

The Hacker News·Aug 22, 08:33 UTC · 25d agoAI safety & security

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·Sep 14, 16:49 UTC · 2d agoAI safety & security

German Manufacturer Shrinks Security Alert Response While Protecting 10,000 Endpoints

Vendor case study: a German manufacturer's five-person SOC cut alert triage time using ANY.RUN's cloud sandbox across 10,000 endpoints.

ANY.RUN published a case study in which a five-person security team at an unnamed German manufacturer replaced an air-gapped forensic laptop with its cloud-managed interactive sandbox, protecting roughly 10,000 endpoints and 10,000 users. The vendor claims a median 15 minutes saved per alert, 20-40 daily tasks processed, a 2.5-minute alert-to-isolation target, and a 95% agreement rate between analyst and sandbox verdicts; all figures are vendor-supplied with the customer identity withheld. The writeup also describes detonating a multi-stage phishing chain from a PDF link to a password-protected ZIP to malware execution.

Cyber Security News·15:12 UTC · 5h agoIndustry 2 sources

Bag of Tricks or Bag of Myths? Reducing Modeling Complexity with Task Knowledge in Explainable Suicide Risk Assessment

Researchers audit 31 NLP techniques for clinician-annotated suicide risk prediction, finding only 5 of 31 comparisons yield reliable gains.

A study of 1,635 clinician-annotated social media posts ran roughly 300 controlled experiments across 7 methodological families, auditing techniques such as model scaling, synthetic data, ensembling, and threshold tuning under severe class imbalance. The proposed system reformulates risk factor prediction as entailment between posts and codebook definitions, using architecturally diverse ensembles with class-balanced training and deployment-consistent calibration. It scores 0.8203 for risk, 0.7953 for evidence, and 0.7045 macro-F1 for factors, ranking third among 53 teams.

arXiv cs.AI / cs.LG / cs.CL·Sep 7, 17:14 UTC · 9d agoAI research

What It’s Like To Run a Tech Giant’s Security Team Without Ever Setting Foot in the Office

The Record·Aug 5, 00:00 UTC · Aug 5, 2026Vulnerability

I-have-ADHD: A skill to stop your coding agent from burying the answer

Developer releases I-have-ADHD, an open-source skill that stops coding agents from burying direct answers under verbose output.

A GitHub project titled I-have-ADHD packages a 'skill' intended to make AI coding agents answer directly instead of hiding the answer inside lengthy output. It gained traction on Hacker News with 45 points and 42 comments. It targets prompt/agent configuration rather than introducing a new model or infrastructure component.

Hacker News · AI·Sep 8, 14:13 UTC · 8d agoAI tools & infraHN 45↑ · 42 comments1

Hacker Conversations: Vinnie Liu, Performer Turned Ringmaster

SecurityWeek interviews Bishop Fox CEO Vinnie Liu, recruited by the NSA at 17 in 1999, on hacker ethics, intent, and his career.

SecurityWeek's Hacker Conversations series profiles Vinnie Liu, who was recruited by the NSA in 1999 at age 17 via an IRC contact and later became CEO of security consulting firm Bishop Fox. The interview covers his white-hat philosophy that hacking for fun differs from hacking to harm, the moral development he attributes to parents and educators, and the industry's shift from the NSA to commercial firms like @stake after its 2000 acquisition of L0pht Heavy Industries. The piece is biographical and opinion-oriented with no incident, vulnerability, or research content.

SecurityWeek·Sep 10, 14:30 UTC · 6d agoIndustry

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·Sep 10, 17:43 UTC · 6d agoAI research

What the Data Says About AI in Security Operations in 2026

Prophet Security's 2026 survey of 250+ security pros: 40% use AI daily; AI users report 25%+ faster investigations and rising AI-driven attacks.

Prophet Security's State of AI in Security Operations 2026 report, based on a survey of 250+ cybersecurity professionals, found 40% of security teams use AI daily and only 4% have no adoption plans. Teams average 100 daily alerts (up to 1,000 at large firms), leave 28% of alerts uninvestigated, and 60% of respondents said missed alerts led to breaches or downtime. Among AI adopters, 72% report at least 25% faster investigations, 56% observed increased AI-driven attacks, and no respondents grant AI full unsupervised autonomy. Data privacy (44%) and explainability (41%) top the adoption hurdles.

The Hacker News·Aug 27, 11:30 UTC · 20d agoIndustry

"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·Sep 7, 21:44 UTC · 8d agoResearch

Can AI make threat intelligence easier? One platform thinks so

Feedly positions its AI-driven Threat Intelligence platform to cut CTI collection time, scanning 10,000+ sources with 1,000 AI models and reporting 70% time savings.

A profile of Feedly Threat Intelligence describes how CTI teams such as RH-ISAC reduced threat data collection from 10 hours weekly to 2-3 hours after adoption. The platform scans over 10,000 open web sources using 1,000 AI models to extract TTPs, CVEs and IoCs into a real-time Threat Graph queryable via an Ask AI LLM with source citations. Customers including GreyNoise, Sopra Steria, gematik and GISA reported measurable gains such as 20 hours saved daily and earlier vulnerability flagging than CISA 65% of the time. The product integrates with Anomali ThreatStream, Cortex XSOAR, Microsoft Sentinel and OpenCTI, plus a REST API and GitHub scripts.

Help Net Security·Aug 24, 04:21 UTC · 23d agoTools

Curriculum Learning as Transport: Understanding Curricula with Wasserstein Geodesics

Researchers model curriculum learning as Wasserstein transport over difficulty distributions, finding curriculum benefits are strongly task- and budget-dependent with no dominant strategy.

The framework represents curricula as trajectories of training distributions over discrete difficulty levels, decoupling ordering, matched exposure, endpoint smoothness, and pacing. Across a calibrated suite of 12 tasks and 33 difficulty axes under fixed training budgets, no single strategy dominates, though easy-to-hard ordering improves hard-level performance relative to exposure-matched static sampling. Endpoint smoothness and pacing substantially affect where along the difficulty spectrum a curriculum is effective, and the transport view supports extensions to learned pacing and structured difficulty spaces.

arXiv cs.AI / cs.LG / cs.CL·Sep 8, 17:39 UTC · 8d agoAI research1

Cyber-Attacks Cost Organizations $52,000 on Average

Hiscox's 2026 survey of 6,800 security leaders found 29% of organizations hit by successful attacks averaging $52,000 in costs and 32.8 hours of downtime.

The Hiscox Cyber Readiness Report 2026, based on a survey of 6,800 security decision-makers across the UK, Europe, and US, found 29% of organizations suffered at least one successful cyber-attack in the past 12 months, averaging four incidents per victim. UK firms were most attacked at 38% while US firms were least at 20%; average incident cost was $52,000 globally, peaking at $134,138 in Italy, with 32.8 hours of average downtime. Impacts included growth delays (32%), financial penalties (28%), and burnout or toxic culture (69%). Businesses invest about $51,000 annually in resilience, and 32% now tie executive compensation to cybersecurity outcomes.

Infosecurity Magazine·11:00 UTC · 9h agoIndustry

Inoculation Midtraining with Learned Neologisms

Inoculation Midtraining confines unsafe LLM behavior to a neologism-marked context, reducing misalignment after unsafe post-training but leaking under nearby contextual cues.

The paper introduces Inoculation Midtraining, which teaches a base model during midtraining that unsafe behavior belongs to a context marked by a learned neologism token, then post-trains on unsafe data within that context. Across supervised fine-tuning and RL post-training regimes, the technique reduces misalignment while preserving transfer of benign properties like German or Shakespearean prose. However, it does not outperform standard Inoculation Prompting, is sensitive to training configuration, and produces a leaky boundary that nearby contextual cues can reactivate. The authors conclude it is not yet a load-bearing component of a developer safety framework.

arXiv cs.AI / cs.LG / cs.CL·Sep 14, 17:12 UTC · 2d agoAI safety & security

From Reweighting to Rewriting: Unlocking the Intervention Effects of Influential Samples in Training Data Attribution

Study shows rewriting responses of influence-selected training examples shifts LLM behavior more strongly than reweighting the same samples.

The paper examines training data attribution, arguing that influence functions identify high-leverage examples whose value goes unrealized under conventional weight-based reweighting interventions. It introduces influence-guided response rewriting, which replaces the responses of influence-selected examples with behavior-aligned or behavior-opposed supervision while keeping instructions fixed, tested across four open-weight LLMs using epistemic abstention as the primary testbed. Rewriting produces stronger, more persistent, and bidirectional behavioral shifts, including on safety refusal, while reweighting the same examples yields weak, inconsistent effects. The results motivate intervention-aware evaluation of TDA methods.

Hugging Face daily papers·Sep 1, 20:00 UTC · 15d agoAI research

What the AI Warning Letter Completely Missed

Opinion piece argues the recent AI warning letter identifies a risk window but omits which actors pose risks and who can mitigate.

This Dark Reading commentary critiques a recent AI warning letter for correctly identifying an approaching risk window while failing to name who is coming through it or who will close it. The piece is brief opinion commentary on AI risk discourse rather than a technical report.

Dark Reading·Sep 3, 17:23 UTC · 13d agoAI safety & security

Ask HN: Anyone still coding like 2021? Where do you work?

Hacker News users debate coding without LLMs, with one developer fired for refusing AI tools and others describing daily hand-coding practice to counter skill atrophy.

An Ask HN thread collects experiences of developers who still write code without LLM assistance. One contributor says he was fired for political reasons after refusing to use LLMs despite adequate stated performance, and observes fewer job ads now require LLM use. Others describe starting each day with a LeetCode problem or 30-60 minutes of hand-coding to stay sharp, contractual bans on AI-generated code for a government-adjacent embedded product over unresolved copyright issues, and inconsistent corporate policies where ChatGPT or Codex use flip-flops between allowed and blocked while a CIO mandates 70-80% AI-generated code next year.

Hacker News · AI·Sep 9, 13:41 UTC · 7d agoAI industryHN 47↑ · 54 comments2

Growing Up The Hard Way

The Hacker News·Aug 7, 11:55 UTC · Aug 7, 2026Vulnerability1

SAFIRE: Safety-Critical Benchmark for Fine-grained Fire and Smoke Understanding in Multimodal LLMs

SAFIRE, an 83K-image fire and smoke benchmark, shows open-source multimodal LLMs average only 61.9% accuracy on safety-critical fire reasoning.

SAFIRE is a large-scale benchmark for fire-smoke understanding in multimodal LLMs with 83K captioned images across 20 scenarios and 193K multiple-choice VQA questions spanning 10 evaluation dimensions from perception to higher-order reasoning. Annotations were built via a GPT-5.4-assisted multi-stage pipeline with MLLM majority voting. Ten open-source MLLMs (8B-38B) average 61.9% accuracy, exposing major gaps in safety-critical reasoning. Adapting vision encoders on 7% of the domain data raises fire-scene classification from 20.1% to 64.5%.

arXiv cs.AI / cs.LG / cs.CL·Sep 7, 17:59 UTC · 9d agoAI research

When will average people feel AI’s impact?

Interconnects essay argues AI's impact is still a rounding error for average people, comparing looming wage stagnation to Engels' pause.

An Interconnects essay argues that AI currently touches daily life far less than previous industrial revolutions, since its benefits are concentrated in knowledge work and lack tangible consumer goods. The author invokes Engels' pause (1790-1840), when British wages stagnated amid rapid GDP growth, as a warning that popular backlash could kneecap AI's development. He contends the current phase is about building compounding infrastructure, and predicts daily life may look similar even 50 years from now.

Interconnects·Sep 9, 11:01 UTC · 7d agoAI industry1

[AINews] 10% worse, 100x cheaper, 10000x faster: Why Simulation is taking over

Latent Space argues AI training pipeline stages—rewards, data, teachers, curricula, environments—are flipping from human-made to model-made simulation.

Latent Space's AINews essay traces how each component of AI training has turned synthetic since 2022: reward models (InstructGPT, RLAIF), synthetic pretraining data (Microsoft Phi, NVIDIA Nemotron-4 340B), model teachers (Alpaca, DeepSeek-R1 distillation), and self-generated curricula (Self-Rewarding Language Models, SPIN). In 2026 it highlights Karpathy's autoresearch loop—700 experiments yielding 20 kept improvements, cutting GPT-2 training time from 2.02 to 1.80 hours—and Z.ai's GLM-5.3 fully synthetic RL environment, judging, and verification stack. It frames these shifts as 'simulation': 10% worse but 100x cheaper and 10,000x faster than human equivalents.

Latent Space·Aug 22, 07:36 UTC · 25d agoAI industry

One Symptom, Three Levers: A Critical Review of On-Policy Self-Distillation

A review paper frames on-policy self-distillation collapse as governed by three levers: token weighting, privileged information, and guidance decay.

The paper critically reviews On-Policy Self-Distillation (OPSD), where a language model trains on its own generations scored token-by-token by a teacher conditioned on privileged information such as reference solutions or environment feedback. It identifies collapse, the progressive narrowing of producible reasoning paths, as the dominant failure mode and analyzes it through three levers: signal weighting, the nature of privileged information, and teacher dynamics. The review is restricted to mathematical reasoning, reports no new experiments, and offers a shared vocabulary separating settled findings from disputed ones.

Hugging Face daily papers·Aug 25, 20:00 UTC · 22d agoAI research

What Matters, When? Diagnosing and Improving Conditional Visual Grounding in Visuomotor Imitation Policies

Researchers diagnose conditional visual grounding failures in visuomotor imitation policies and show targeted interventions substantially improve distractor robustness.

The paper studies why ACT-based visuomotor imitation policies fail when visually similar distractor objects or receptacles are introduced, finding sensitivity depends on both distractor type and manipulation stage. Interventions including distractor augmentation, phase-dependent attention regularization, and appearance-based visual prompting improve target selection while preserving spatial control information, with gains in simulation and on a physical UR3e. The same failure pattern is confirmed in a pretrained vision-language-action policy on a state-conditioned medical instrument-handling task.

arXiv cs.AI / cs.LG / cs.CL·Sep 4, 17:27 UTC · 12d agoAI research

Measuring LLM Sycophancy under Sustained Multi-Turn Pressure

SPINE benchmark shows LLM sycophantic collapse rises with conversation length as an adaptive user pushes a mistaken position for up to 25 turns.

The SPINE benchmark uses an LLM proxy that persistently and adaptively defends a mistaken user position for up to 25 turns, testing four production LLM systems and three OLMo3-7B variants on 100 false-presupposition and 100 unethical-query items. Collapse rates increase with conversation length for every model, and short-horizon evaluation protocols underestimate sycophancy. Analysis of accessible reasoning traces shows the correct position often remains represented when the model concedes, indicating models choose to please users rather than lacking knowledge. Among tested tactics, emotional appeals are most associated with inducing sycophantic behavior.

arXiv cs.AI / cs.LG / cs.CL·Sep 8, 17:35 UTC · 8d agoAI safety & security1

WearableQA: A Benchmark for Health Reasoning over Real-World Wearable Data

WearableQA benchmark introduces 4,084 questions over real longitudinal wearable data, showing 14 LLMs score 19.6-72.9% on health reasoning, far from solved.

WearableQA is a benchmark of 4,084 10-option multiple-choice questions built from wearable time series, blood biomarkers, and demographics of 200 real users with up to 500 days of daily measurements. It defines 16 question types along two axes: data versus health reasoning, and single- versus cross-signal reasoning, using a dual-grounding framework combining literature and population-validated patterns. Evaluations of 14 proprietary and open-source LLMs show accuracy ranging from 19.6% to 72.9% against a 10% chance baseline, with most models below 60%.

Hugging Face daily papers·Sep 3, 20:00 UTC · 13d agoAI research

Show HN: LLM Attention Visualization

A developer released a browser-based tool that visualizes which past tokens influence each LLM output token using aggregated, value-weighted attention scores.

A Show HN project presents a React application built on Transformers.js that renders per-token attention influence by aggregating attention weights scaled by value-vector magnitudes across all attention heads and layers. To expose internal tensors, the author instrumented the ONNX computation graph, hosted a modified model on Hugging Face, and pre-generated prompts to avoid long model downloads in the browser. Demos with a 600-million-parameter model show how verbatim copying draws heavily on source tokens and how single outputs blend information from multiple phrases.

Hacker News · AI·Sep 8, 16:59 UTC · 8d agoAI tools & infraHN 41↑ · 9 comments

Necessary or Sufficient? Evaluating LLM Explanations With Behavioural Evidence

New framework tests whether LLM-cited explanation factors are necessary or sufficient, finding weak correlation across Claude, GPT, and Gemini models.

An arXiv paper introduces black-box intervention tests measuring whether factors LLMs cite in their explanations are necessary or sufficient for their outputs in agent oversight workflows. Across eight models from the Claude, GPT, and Gemini families, Spearman correlations between cited rankings and measured influence ranged from 0.349-0.354 (advisor recommendation) to 0.431-0.580 (prompt monitoring). Uncited factors scored above the lowest cited factor in up to 57.6% of advisor responses, showing cited top-three factors do not reliably identify the most influential inputs.

arXiv cs.AI / cs.LG / cs.CL·Sep 4, 17:37 UTC · 12d agoAI safety & security1

ActReview: Rebuttal-Guided Training Data and Rubric Rewards for Actionable Peer Review Generation

ActReview post-trains Qwen3-8B-Base on OpenReview rebuttals to generate actionable peer-review feedback with grounded revision suggestions, benchmarked on 1,000 curated instances.

The paper defines Actionable Peer-review Generation as diagnostic claim generation plus revision suggestion generation and introduces ActReview, a rebuttal-guided post-training framework. From OpenReview review-rebuttal threads the authors build ActReview-40K, aligning reviewer weaknesses with author responses grounded in localized paper evidence, and post-train Qwen3-8B-Base with multi-task SFT followed by GRPO using weakness-specific rubric rewards. They also release ActReview-Bench, a human-curated 1,000-instance benchmark, on which ActReview outperforms prior specialized review-generation models on actionability and grounding while remaining competitive with strong prompt-based LLMs. Human evaluation confirms improved revision usefulness but identifies a remaining gap in technical accuracy.

Hugging Face daily papers·Sep 7, 20:00 UTC · 9d agoAI research

Better answers, broader thinking: What students gain from ChatGPT and critical-thinking training

OpenAI published a randomized study of over 1,000 students examining ChatGPT's effects on critical thinking and performance.

OpenAI released results from a randomized study of more than 1,000 students using ChatGPT on a real-world university assignment. The research examined effects on critical thinking, originality, and student performance. Findings inform ongoing debates about AI's role in education and learning outcomes.

OpenAI News·Aug 27, 09:00 UTC · 20d agoAI research

A Chosen Future Can Still Be Rewritten: Causal Writability in Video Models

Study shows video models often learn correct physics but fail to use it; low-dimensional 'causal writability' edits can restore correct motion.

The paper demonstrates 'causal writability' in video generation models: physically correct motion remains available inside the model even when the model outputs incorrect motion. In a red/blue mass oscillation setup, a low-dimensional edit predicted from simple physical variables restores correct fast motion, with a sharp depth boundary marking commitment. Early causal writability predicts which training errors later get corrected, and both writability and closure reproduce in a pretrained 1.3B video model.

arXiv cs.AI / cs.LG / cs.CL·Sep 14, 17:58 UTC · 2d agoAI research

Mythos Vulnerability Firehose Hits a Human Bottleneck

Analysis of Project Glasswing findings shows only a fraction of discovered vulnerabilities have reached disclosure and even fewer are fixed.

Dark Reading reports that an analysis of Project Glasswing findings shows only a fraction of the vulnerabilities discovered by the program have reached disclosure, and an even smaller number have been fixed. The article examines how the volume of findings from the discovery program is bottlenecked by limited human triage and remediation capacity. The piece highlights growing tension between high-volume vulnerability discovery and the industry's ability to process, disclose and patch reports.

Dark Reading·Sep 9, 21:19 UTC · 6d agoResearch

Causal Foundation Models

A paper introduces causal foundation models (CFMs): pretrained networks that estimate treatment effects on new datasets via in-context learning without fine-tuning.

Causal foundation models (CFMs) apply the foundation-model paradigm to causal inference, replacing bespoke per-problem estimator pipelines with networks pretrained once at scale. CFMs estimate causal quantities such as the average treatment effect on entirely new datasets through in-context learning, without model updates. The work serves as a practical introduction to the emerging area, covering background in causal inference and machine learning and including example code and Jupyter notebooks.

Hugging Face daily papers·Sep 1, 20:00 UTC · 15d agoAI research