# DevFeed -- Developer news Published developer news, articles, tutorials, releases, comparisons and opinions. This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles. ## Meta ordered to pay $25 billion over Cambridge Analytica privacy case DevFeed: [Meta ordered to pay $25 billion over Cambridge Analytica privacy case]() Original publisher: [Read original article]() Author: Edward Komenda Published: 2026-10-22T09:20:08.594854Z Content type: article Language: en Sources: [The Proton Blog]() Topics: [online privacy](), [Security & Privacy]() Tags: [ads](), [ai](), [cambridge-analytica](), [data](), [facebook](), [meta](), [new-mexico](), [news](), [privacy]() ### AI overview A New Mexico judge ordered Meta to pay $25 billion plus 15% annual interest after a jury found more than 40 million violations of the state's Unfair Practices Act. The case centered on Facebook telling users they controlled their data while not explaining third-party access, including the Cambridge Analytica episode. Meta says the state did not show consumer harm and plans to appeal, which could take years. The penalty does not change Facebook's current data practices, and New Mexico law directs the funds to public education rather than affected users. The excerpt also describes other reported or alleged Meta privacy and platform controversies, while qualifying that Meta denied one allegation. ### Source excerpt A New Mexico judge ordered Meta to pay $25 billion for misleading Facebook users about their data. What the Cambridge Analytica ruling means. ## Authorities disrupt nearly 11,000 cybercrime networks in Africa DevFeed: [Authorities disrupt nearly 11,000 cybercrime networks in Africa]() Original publisher: [Read original article]() Author: Bill Toulas Published: 2026-10-22T08:27:11Z Content type: news Language: en Sources: [BleepingComputer]() Topics: [Cybercrime](), [Security Attacks](), [online-fraud-prevention](), [ransomware](), [Social engineering]() Tags: [2026](), [africa](), [business-email-compromise](), [command-and-control](), [computer-help](), [computer-security](), [computers](), [credentials](), [cryptocurrency](), [cryptocurrency-investment](), [cybercrime](), [cybercriminal](), [cybersecurity](), [disrupt](), [fraud](), [illegal](), [infosec](), [infosec-computer-security](), [intelligence](), [interpol](), [legal](), [linux](), [mac](), [malware](), [malware-removal](), [ransomware](), [scams](), [security](), [spyware](), [support](), [tech-support](), [technical-support](), [virus](), [virus-removal](), [windows]() ### AI overview Operation Serengeti III, coordinated by Interpol across 21 countries from June 29 to August 31, led to 1,462 arrests, the dismantling of 10,909 malicious networks, and recovery of more than $105 million. Investigators identified 65,929 victims and estimated losses above $385 million. The operation targeted business email compromise, ransomware and online scams, among other crimes. Examples include fraud arrests and ring takedowns in Uganda, seized cryptocurrency mining equipment in Angola, and a Nigerian suspect linked to phishing and business email compromise. Private partners including Team Cymru, Kaspersky, Group-IB and Fortinet provided intelligence and technical expertise, such as information on suspicious domains, IP addresses and command-and-control infrastructure, to help map networks and identify suspects. ### Source excerpt Law enforcement agencies have disrupted nearly 11,000 cybercriminal networks and arrested more than 1,400 people in Operation Serengeti III, which targeted business email compromise, ransomware, and online scams. [...] ## Africa cybercrime crackdown nets 1,462 arrests and over $105 million DevFeed: [Africa cybercrime crackdown nets 1,462 arrests and over $105 million]() Original publisher: [Read original article]() Author: Anamarija Pogorelec Published: 2026-10-22T08:03:06Z Content type: news Language: en Sources: [Help Net Security]() Topics: [Cybercrime](), [online-fraud-prevention]() Tags: [2026](), [africa](), [arrest](), [arrested](), [asia](), [business-email-compromise](), [command-and-control](), [credentials](), [cryptocurrency](), [cybercrime](), [fortinet](), [group-ib](), [illegal](), [interpol](), [kaspersky](), [law-enforcement](), [malicious](), [meta](), [networks](), [news](), [phishing](), [ransomware](), [scams](), [team-cymru]() ### AI overview Operation Serengeti III, an INTERPOL-coordinated sweep across 21 countries from June 29 to August 31, 2026, resulted in 1,462 arrests and the disruption of 10,909 malicious networks. Investigators recovered more than $105 million and linked the networks to over $385 million in losses and 65,929 victims. INTERPOL supplied intelligence on domains, IP addresses and command-and-control infrastructure, while private firms helped identify targets and map networks. Cases included fraud call centers dismantled in Uganda, illegal crypto-mining facilities shut in Angola, phishing and investment fraud investigations in Nigeria, and schemes disrupted in Tanzania, Gabon and the Republic of Congo. ### Source excerpt Police in 21 countries arrested 1,462 suspects and took down 10,909 malicious networks in Operation Serengeti III, an INTERPOL-coordinated cybercrime sweep across Africa that ran from June 29 to August 31, 2026. Investigators recovered more than $105 million in illicit funds. Financial investigators tied the networks to losses exceeding $385 million and identified 65,929 victims of fraud, ransomware, business email compromise, online scams and digital extortion. Twenty of the countries were African, and the 21st ... More -> The post Africa cybercrime crackdown nets 1,462 arrests and over $105 million appeared first on Help Net Security. ## Torin Deep Dive: Lazy Evals in GPU-Native SQL Engine DevFeed: [Torin Deep Dive: Lazy Evals in GPU-Native SQL Engine]() Original publisher: [Read original article]() Published: 2026-10-22T08:00:33Z Content type: article Language: en Sources: [Starburst Data]() Topics: [SQL](), [GPU rendering]() Tags: [cudf](), [engine](), [gpu](), [lazy](), [sql](), [starburst]() ### AI overview Torin preserves SQL's short-circuit behavior for fallible expressions while keeping safe filters on its fast, column-at-a-time GPU path. It classifies expressions by how they can fail: eager evaluation for infallible expressions; NULL masking when NULL inputs prevent failure; and filter-and-scatter when masking cannot make evaluation safe, at the cost of copying active rows and restoring results. Torin routes lazy constructs such as AND, OR, COALESCE, IF and CASE through a shared IF mechanism, using LET to avoid recomputing values. It rewrites only where needed, so comparison-only filters remain eager. The text describes the design and tradeoffs but provides no performance measurements for the strategies. ### Source excerpt Torin is the new GPU execution engine we built at Starburst. It runs SQL query ... ## Trying to Get Linux Running on a Nokia E6 DevFeed: [Trying to Get Linux Running on a Nokia E6]() Original publisher: [Read original article]() Author: Zoe Skyforest Published: 2026-10-22T08:00:17Z Content type: news Language: en Sources: [Hackaday]() Topics: [Linux](), [смартфоны]() Tags: [app](), [boot](), [bootloader](), [broadcom](), [chip](), [custom](), [device-tree](), [display](), [driver](), [functional](), [hack](), [hacks](), [hardware](), [home-lab](), [issue](), [kernel](), [linux](), [linux-hacks](), [nokia](), [nokia-e6](), [phone-hacks](), [smartphone]() ### AI overview Without an unlocked bootloader, the Nokia E6 boots Linux through a Symbian app that loads the kernel and device tree into RAM, disables the MMU and caches, then jumps to Linux; returning to Symbian requires a power cycle. The phone provides a terminal, with its keyboard and touchscreen working. A custom display driver replays messages captured from Nokia's original driver for the Broadcom VideoCore III. Linux always resets within a few minutes. The developer suspects the baseband resets the application processor after detecting that Symbian's mobile stack has stopped, but this remains unconfirmed. ### Source excerpt The Nokia E6 is a smartphone from back in the old days when Symbian was a thing. [fluffleswaffled] has been doing the work to get the ancient handset running Linux ...read more ## Using External Tools in BGP Labs DevFeed: [Using External Tools in BGP Labs]() Original publisher: [Read original article]() Published: 2026-10-22T07:51:00Z Content type: article Language: en Sources: [ipSpace]() Topics: [BGP](), [Tool]() Tags: [bgp](), [codespaces](), [external](), [labs](), [netlab](), [tools]() ### AI overview The document explains how to start external tools, including network-xray, in open-source BGP configuration labs without changing the lab topology. It also describes how this works in GitHub Codespaces and says lab instructions now point readers to an optional tool setup step. ### Source excerpt After Kazuya Takahashi released his wonderful network-xray tool and added it as an external tool to the open-source BGP configuration labs, I wrote a short document explaining how to start external tools in your BGP labs without changing the lab topology, and how it works in GitHub Codespaces (hint: wonderfully). I also modified all lab instructions to include a pointer to the optional tool-setup step. Happy labbing ;) ## PVTW: Extending Silicon Characterization From PVT To Real-World Workload DevFeed: [PVTW: Extending Silicon Characterization From PVT To Real-World Workload]() Original publisher: [Read original article]() Author: Gerard John Published: 2026-10-22T07:07:27Z Content type: article Language: en Sources: [Semiconductor Engineering]() Topics: [Chip design]() Tags: [advanced-packaging](), [ai](), [ai-accelerator](), [ai-processor](), [amkor](), [application](), [architecture](), [ate](), [behavior](), [blogs-md](), [boot](), [characterization](), [chip](), [chip-package-system](), [clock](), [clock-distribution](), [clocks](), [combined](), [manufacturing-packaging-and-materials](), [pvt](), [pvtw](), [real-world](), [silicon](), [slt](), [system-level-test](), [temperature]() ### AI overview PVTW extends silicon characterization from Process, Voltage, and Temperature to include Workload, since devices at identical electrical conditions can behave differently under different workloads. The proposed hybrid approach combines ATE's controlled power, rail and clock adjustment, structural patterns, parametric measurement, and SHMOO capability with SLT's ability to boot software and run deterministic system workloads. It supports ATE, SLT, and combined modes; the combined mode sweeps electrical conditions while workloads exercise compute, memory, or I/O subsystems. The excerpt illustrates how differing voltage margins under ATPG and an SRAM-intensive workload can reveal workload-dependent margin consumption. It stresses that characterization requires controlled, repeatable conditions and deliberate boundary sweeps, and presents a loop for reproducing marginal behavior, isolating subsystems, and correlating workload and electrical results. ### Source excerpt Two devices operating at exactly the same voltage, frequency, and temperature can behave differently depending on what they are being asked to do. The post PVTW: Extending Silicon Characterization From PVT To Real-World Workload appeared first on Semiconductor Engineering. ## Ruby 4.0.8 Released DevFeed: [Ruby 4.0.8 Released]() Original publisher: [Read original article]() Published: 2026-10-22T06:54:51Z Content type: release Language: en Sources: [Ruby-lang]() Topics: [Ruby](), [releases]() Tags: [2026](), [bugfixes](), [details](), [release](), [ruby]() ### AI overview Ruby 4.0.8 is a routine update containing bug fixes; the supplied excerpt does not identify the fixes. It lists gzip, xz, and zip downloads with file sizes and SHA-1, SHA-256, and SHA-512 checksums. The release credits contributors who reported bugs and helped prepare it. ### Source excerpt Ruby 4.0.8 has been released. This is a routine update that includes bugfixes. Please see the GitHub releases for further details. Download https://cache.ruby-lang.org/pub/ruby/4.0/ruby-4.0.8.tar.gz SIZE: 23947459 SHA1: f79bc8217dd0edd7ecc0c1c0cb57e64a906258a5 SHA256: 0968e2ac4cbd6f1fd85d3092aed66cd3c2d8ba4a5e2493d6e6c6cb0c431897d2 SHA512: 011b741c8fe3160643739f8fd88f3a69b198260dabcd3cb791c1ffacdfae7b48f6b2ca365ea97198d5dcb13fc839533ec1e83bab76f38c25203a7a3b6e665e8b https://cache.ruby-lang.org/pub/ruby/4.0/ruby-4.0.8.tar.xz SIZE: 18019152 SHA1: 24404aa1371a1f5a28bfcc8ac4331bd744810954 SHA256: 3bd35abc39dce9c682d62cf333fbd8b8f34816dd651f8228b3b59353157b84a4 SHA512: 79eb57f30a853d6e856ce2df35b8ff20c393d5cacad43420c5caba9ebba53445d1b7b40f0a65a1cd4eee6bfc86abc3eb722e1fa5bef10b41bf56779b566e894a https://cache.ruby-lang.org/pub/ruby/4.0/ruby-4.0.8.zip SIZE: 29263079 SHA1: 40c0e4f791563bc8a24d683547cb3fe08c01da4b SHA256: 68e12bb40c9175c2a97f371ff6a76dd93649bb09dc76d757394b7086c881ade4 SHA512: bee8b619c287530d10779d6527d5e7397b75b74d0abe57d9a12b8d8d0b303064ea6da3fe9a47d1fb9835abfe11701e64a767841658e209cb866ca44aa951ef33 Release Comment Many committers, developers, and users who provided bug reports helped us make this release. Thanks for their contributions. Posted by k0kubun on 22 Oct 2026 ## AI-found bugs are expected to rise, and OpenAI has a five-stage response plan DevFeed: [AI-found bugs are expected to rise, and OpenAI has a five-stage response plan]() Original publisher: [Read original article]() Author: Mirko Zorz Published: 2026-10-22T07:35:52Z Content type: news Language: en Sources: [Help Net Security]() Topics: [Security research](), [bugfixes](), [Open Source](), [OpenAI]() Tags: [ai](), [applications](), [bugs](), [calif](), [critical](), [cybersecurity](), [gpt](), [gpt-5-6-sol](), [model](), [news](), [open-source](), [open-source-security](), [openai](), [patch](), [patch-the-planet](), [php](), [plan](), [process](), [rise](), [security](), [software](), [stage](), [vulnerability-disclosure]() ### AI overview OpenAI proposes a five-stage response for critical vulnerabilities affecting widely used software, based on a two-month effort around PHP flaws found by Calif using GPT-5.6 Sol and independently reported by Wiz. Maintainers lead confirmation, impact assessment, patching, and testing, including using frontier models to probe fixes and find variants. Responders then give affected distributors and service providers time to prepare, engage government, broaden coordination once plans settle, and publish guidance while operators deploy patches. Early access depends on an organization's protective role, response time needed, and ability to handle sensitive details; maintainers decide what each participant receives. Teams may revise timelines if exploitation, leaks, revealing patches, or delays increase risk. OpenAI seeks input on disclosure windows, leak risks, unsupported package versions, and whether authorized parties should patch systems after losing administrative access. ### Source excerpt OpenAI has proposed a five-stage process for handling critical vulnerabilities in widely used software. The process grew out of a two-month response to flaws in PHP, the open-source language that runs websites and web applications. Calif found one of the PHP flaws using OpenAI's GPT-5.6 Sol model, as part of OpenAI's Patch the Planet initiative for open-source security, and Wiz independently found and reported both. "As significant as they are, these bugs are among several ... More -> The post AI-found bugs are expected to rise, and OpenAI has a five-stage response plan appeared first on Help Net Security. ## Stacked-Die Multi-Chip Architectures: The Tradeoff Between Scaling And Power Delivery DevFeed: [Stacked-Die Multi-Chip Architectures: The Tradeoff Between Scaling And Power Delivery]() Original publisher: [Read original article]() Author: Kaladhar Radhakrishnam Published: 2026-10-22T07:02:07Z Content type: article Language: en Sources: [Semiconductor Engineering]() Topics: [Chip design]() Tags: [architectures](), [bandwidth](), [blogs-md](), [capacitor](), [decoupling](), [delivery](), [foundry-focus](), [integrated-voltage-regulators](), [intel](), [intel-foundry](), [manufacturing-packaging-and-materials](), [multi-chip-assemblies](), [networks](), [power](), [power-delivery](), [regulation](), [regulators](), [scaling](), [semiconductor](), [stacked-die](), [voltage-droop](), [voltage-regulators]() ### AI overview As stacked-die and multi-chip systems add compute resources, rising current density, more voltage domains, routing congestion, and bursty workloads make voltage droop and power integrity harder to manage. The excerpt describes moving voltage conversion closer to the load as a way to reduce distribution losses and improve response. Intel's FIVR with CoaxMIL uses magnetic inductors and is presented as a higher-current-density, smaller-area approach; C2VR uses capacitors integrated onto a regulator chiplet for compact conversion and rapid response; IVR chiplets offer placement flexibility, including within package layers or beneath compute dies, and may enable vertical power delivery. The approaches address different design needs, while the excerpt gives no quantitative comparison of their performance. ### Source excerpt Integrated voltage regulators reduce current density, provide higher regulation bandwidth, and reduce reliance on large decoupling capacitor networks. The post Stacked-Die Multi-Chip Architectures: The Tradeoff Between Scaling And Power Delivery appeared first on Semiconductor Engineering. ## Datadog vs. Dynatrace: An Honest Comparison for Cloud-Native Teams (2026) DevFeed: [Datadog vs. Dynatrace: An Honest Comparison for Cloud-Native Teams (2026)]() Original publisher: [Read original article]() Author: John Blust Published: 2026-10-22T06:53:19.552941Z Content type: comparison Language: en Sources: [New Relic]() Topics: [datadog](), [dynatrace](), [OpenTelemetry]() Tags: [2026](), [ai](), [analysis](), [apm](), [application-monitoring](), [architecture](), [cloud-native](), [comparison](), [datadog](), [deployment](), [devops](), [dynatrace](), [pricing](), [root-cause-analysis]() ### AI overview Datadog's modular, product-by-product setup offers teams selective coverage but adds configuration and separate usage charges; Dynatrace's OneAgent automatically discovers and maps instrumented systems in Smartscape, speeding initial setup while requiring enterprise governance at scale. Their AI tools, Watchdog and Davis AI, identify likely causes using the telemetry each can access, so gaps in purchased or instrumented coverage limit both. Both platforms use usage-based pricing that can be difficult to forecast: Datadog stacks product-specific meters, while Dynatrace draws several usage categories against an annual commitment and says it has no penalty-style overages. The article also describes New Relic as an alternative with one data model, native OpenTelemetry support, ingestion-based pricing, and AI correlation across metrics, traces, and logs. These comparisons reflect the supplied article excerpt. ### Source excerpt Comparing Datadog and Dynatrace on architecture, deployment, AI-driven root cause analysis, and pricing -- plus where a unified alternative fits in. ## NVIDIA Cosmos Opens New Tools to Train, Evaluate and Deploy World Models DevFeed: [NVIDIA Cosmos Opens New Tools to Train, Evaluate and Deploy World Models]() Original publisher: [Read original article]() Author: Pranjali Joshi Published: 2026-10-22T05:00:27Z Content type: release Language: en Sources: [NVIDIA]() Topics: [Cosmos](), [Metropolis](), [NVIDIA NIM](), [Robotics Simulation](), [World models](), [AI Development]() Tags: [agent-skills](), [agentic](), [ai](), [autonomous](), [autonomous-vehicles](), [azure](), [blueprints](), [cloud-native](), [computer-vision](), [cosmos](), [data](), [data-augmentation](), [dataset](), [deploy](), [factory](), [gtc-2026](), [physical-ai](), [simulation](), [train](), [training](), [world-models]() ### AI overview NVIDIA's Cosmos platform combines data preparation, model training and optimization, simulation, and deployment tools for physical AI. Cosmos Data Factory offers five cloud-native blueprints for curating, augmenting, and generating image or video data; Cosmos Curator now supports multimodal, multisensory inputs. Developers can fine-tune and use reinforcement learning with compatible open models, then apply distillation and quantization. New Cosmos 3 Nano simulation variants support repeatable testing of autonomous policies, while updated Cosmos 3 NIM microservices report up to 2x faster inference with no quality loss in validated configurations. Reported uses include Porsche's shared AV data workflow, with projected reductions of up to 90% in repeated test drives and 85% in development cycle time, and ANYbotics' generation of over 1,500 synthetic inspection clips. These outcomes are examples from the excerpt, not guarantees for other deployments. ### Source excerpt To help developers rapidly build world models, which help physical AI systems understand and simulate the real world, NVIDIA has expanded the Cosmos platform with new training tools, simulation checkpoints and Cosmos Data Factory blueprints. Ranking top five on six different open models leaderboards, the NVIDIA Cosmos platform gives world model builders the frontier open [...] ## NeuroCodec: EEG-Guided Latent-Domain Generative Model for High-Fidelity Target Speaker Extraction DevFeed: [NeuroCodec: EEG-Guided Latent-Domain Generative Model for High-Fidelity Target Speaker Extraction]() Original publisher: [Read original article]() Author: Julian Silva, Jaliya Kodithuwakku, Navindu Sirimanna, Shaveen Herath, Wageesha N. Manamperi, Joshua P. Kulasingham Published: 2026-10-22T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Transformer](), [generative-ai-tools](), [Mamba](), [Generative Adversarial Network](), [fourier transform]() Tags: [adversarial](), [attention](), [audio](), [codec](), [cs-lg](), [datasets](), [eeg](), [eess-as](), [generative](), [mamba](), [masking](), [metrics](), [neural](), [objectives](), [party](), [processing](), [real-time](), [regions](), [space](), [spatial](), [speaker](), [target]() ### AI overview NeuroCodec uses EEG to guide target speaker extraction by generating speech in a neural audio codec's latent space, addressing fidelity limits of time-frequency masking in overlapping speech and narrowband audio. It models spatial EEG dependencies with graph convolutions and combines EEG and speech representations using attention and Mamba-based temporal modeling. A six-phase curriculum adds latent reconstruction, multi-scale spectral, and adversarial objectives. On the Cocktail Party and KUL datasets, it outperformed baselines on several perceptual metrics; in a 20-person listening study, its mean score was 82.24 versus 49.63 for M3ANet. The abstract reports 44.1 kHz reconstruction with spectral content up to 20 kHz and a causal streaming formulation for real-time processing. ### Source excerpt arXiv:2610.29815v1 Announce Type: cross Abstract: Driven by recent advancements in hearing assistive devices, target speaker extraction (TSE) in multi-talker environments has drawn significant attention. To address this, hearing technologies identify attention using neural signals recorded via electroencephalography (EEG) to guide TSE, enabling neuro-steered auditory attention decoding. However, most existing techniques rely on time-frequency masking, which suffers from low fidelity in spectro-temporally overlapping regions and is often restricted to narrowband audio. In contrast, we introduce NeuroCodec, which, to the best of our knowledge, is the first generative EEG-guided TSE framework operating in the latent space of a neural audio codec. Specifically, NeuroCodec models spatial EEG dependencies via graph convolutions and fuses cross-modal EEG and speech representations through attention mechanisms and Mamba-based temporal modeling. We further introduce a six-phase training curriculum that progressively incorporates latent reconstruction, multi-scale spectral, and adversarial objectives to ensure stable, high-fidelity speech generation. Experiments on the public Cocktail Party and KUL datasets show NeuroCodec outperforms baselines, across several perceptual metrics. A listening experiment with 20 participants validated that NeuroCodec offers greater perceptual accuracy, achieving a mean score of 82.24 compared with 49.63 for M3ANet. Additionally, the proposed method is seen to enable 44.1 kHz reconstruction with spectral content extending up to 20 kHz, further supporting the perceptual results. Finally, we extend NeuroCodec with a causal streaming formulation suited for real-time processing, providing a basis for hearing assistive devices. ## Motivation and Contribution in Open Source Software: A Meta-Analytic Investigation DevFeed: [Motivation and Contribution in Open Source Software: A Meta-Analytic Investigation]() Original publisher: [Read original article]() Author: Parisa Omidmand, Saeed Nosratabadi Published: 2026-10-22T04:00:00Z Content type: article Language: en Sources: [arXiv: Software Engineering]() Topics: [Open Source](), [intrinsic motivation]() Tags: [contribution](), [cs-hc](), [cs-se](), [econ-gn](), [meta](), [motivation](), [open-source](), [q-fin-ec]() ### AI overview A meta-analysis of 41 studies finds that motivations relate more strongly to contributors' intention to continue than to their actual effort. The main analysis pools 178 correlations covering eight antecedents across 29 independent samples (N=9,632): seven antecedents are positively related to effort, all eight to continuance intention, and each relates at least as strongly to intention. Intrinsic motivation is not more strongly related to effort than extrinsic motivation, though its intention-effort gap is larger. Reputation relates to effort mainly on platforms with reputation systems; the analysis finds no evidence that monetary rewards crowd out enjoyment. ### Source excerpt arXiv:2610.29199v1 Announce Type: cross Abstract: Research on why people contribute to open source software (OSS) and related open collaboration platforms spans more than two decades, but its findings conflict, and studies often treat the work people contribute and their intention to keep contributing as interchangeable. Drawing on self-determination theory and motivation crowding theory, we meta-analyze 41 studies. The main analysis pools 178 correlations of eight antecedents with contribution effort and continuance intention from 29 independent samples (N = 9,632). Seven antecedents relate positively to effort, all eight relate positively to intention, and each relates at least as strongly to intention as to effort. Contrary to a common assumption, intrinsic motivation is not more strongly related to effort than extrinsic motivation, but its gap between intention and effort is larger. Reputation relates to effort mainly on platforms with a reputation system, and we find no sign that monetary rewards crowd out enjoyment. The findings inform how platforms can turn contributors' intentions into work. ## Closing the Sim-to-Real Gap: An Evaluation Framework for Autonomous Cyber Defense Configuration of Commercial EDR DevFeed: [Closing the Sim-to-Real Gap: An Evaluation Framework for Autonomous Cyber Defense Configuration of Commercial EDR]() Original publisher: [Read original article]() Author: Kerri Prinos, Lilianne Brush Published: 2026-10-22T04:00:00Z Content type: article Language: en Sources: [arXiv: Cryptography and Security]() Topics: [Endpoint Security & XDR](), [Agentic SOC]() Tags: [active-directory](), [agent](), [agents](), [ai](), [autonomous](), [benchmark](), [configuration](), [crowdstrike](), [cs-ai](), [cs-cr](), [cyber](), [cyber-defense](), [defense](), [edr](), [endpoint-detection-and-response](), [enterprise](), [evaluation](), [first](), [framework](), [gap](), [hardening](), [language-model]() ### AI overview The framework evaluates autonomous defense agents hardening commercial EDR in a setting where the EDR itself makes vendor-specific autonomous decisions. It is instantiated in a Game of Active Directory lab with Horizon3's NodeZero as the pentester and Microsoft Defender XDR as the primary EDR, then extended to CrowdStrike Falcon and SentinelOne Singularity. A sample benchmark uses Claude Sonnet 4.6 and Cisco Foundation-Sec-8B. The authors identify three evaluation challenges: commercial EDR telemetry is designed for SOC workflows rather than measuring policy effectiveness; per-policy attribution is needed to distinguish agent actions from EDR actions; and results vary across runs, largely because containment varies and accumulates. The abstract motivates methods that account for these black-box behaviors when benchmarking enterprise defense agents. ### Source excerpt arXiv:2606.08168v2 Announce Type: replace Abstract: Leading commercial endpoint detection and response (EDR) products have shifted from operator-configured rule sets to multi-component systems where autonomous AI components operate alongside, and increasingly in place of, operator-deployed policies. Autonomous defense agents using commercial EDR as their hardening tool are no longer tuning a passive tool, but a black-box autonomous system capable of making vendor-specific decisions. We present the first evaluation framework for autonomous defense agents hardening commercial EDR. We instantiate it in a Game of Active Directory (GOAD) lab with Horizon3's NodeZero as the autonomous pentester and Microsoft Defender XDR as the primary EDR under test. We extend the same framework to CrowdStrike Falcon and SentinelOne Singularity to test whether our Microsoft Defender XDR findings hold across vendors. We run a sample benchmark of defense agents with two large language model (LLM) backbones (Claude Sonnet 4.6 and Cisco Foundation-Sec-8B). We report three lessons learned that neither simulation nor open source EDR evaluation can surface: (i) commercial EDR telemetry is engineered for Security Operations Center (SOC) analyst workflows rather than for measuring policy effectiveness; (ii) per-policy attribution is important for separating the defense agent's actions from the EDR's autonomous actions; and (iii) EDR behavior varies between runs, largely because containment varies and accumulates. Together, these findings highlight a sim-to-real gap for enterprise defense and motivate evaluation methodology for benchmarking autonomous defense agents in environments with black-box, autonomous tools. ## Through the Looking Glass: Analyzing Geolocation Providers Using Looking Glasses and Active Measurements DevFeed: [Through the Looking Glass: Analyzing Geolocation Providers Using Looking Glasses and Active Measurements]() Original publisher: [Read original article]() Author: William Leung, Oliver Gasser Published: 2026-10-22T04:00:00Z Content type: article Language: en Sources: [arXiv: Networking]() Topics: [Internet Traffic](), [communications]() Tags: [cs-ni](), [geolocation](), [measurements]() ### AI overview The study evaluates 4,487 Looking Glass servers across 144 countries by comparing their locations in six geolocation databases from four providers and checking them with active measurements. In round-trip-time measurements, 9 of 18 Looking Glass providers had at least one server whose location was disconfirmed by the RTT evidence. The databases differed substantially in how closely they located the servers; IPinfo showed the closest distances. These findings indicate that both provider choice and geolocation database affect how reliably Looking Glasses can serve as measurement vantage points. ### Source excerpt arXiv:2610.29660v1 Announce Type: new Abstract: Looking Glass servers (LGs) allow users and network operators to debug issues or measure performance to/from arbitrary, external IP addresses, making LGs a valuable tool in a variety of analyses. To assess the reliability of LGs as an independent and free set of vantage points, we perform an analysis of 4487 LG servers across 144 countries. We analyze LGs by comparing them to six different geolocation databases from four providers, and by conducting active measurements on LGs. In our RTT measurements we find 9 of 18 LG providers have at least one server with an RTT-disconfirmed location. We also find substantial differences between geolocation databases' ability to locate LGs, with IPinfo showing the closest distance to LGs. ## NVIDIA Transforms European Retail, From Digital Discovery to Physical Operations DevFeed: [NVIDIA Transforms European Retail, From Digital Discovery to Physical Operations]() Original publisher: [Read original article]() Author: Mavi Nafria Published: 2026-10-22T05:00:23Z Content type: news Language: en Sources: [NVIDIA]() Topics: [Physical AI](), [Omniverse](), [Robotics Simulation](), [3D](), [Nemotron](), [NeMo](), [CUDA](), [TensorRT](), [RTX PRO](), [digital sovereignty](), [Image](), [Nvidia]() Tags: [accelerated](), [ai](), [collaboration](), [commerce](), [costs](), [digital](), [digital-twin](), [discovery](), [europe](), [gtc-2026](), [inference](), [inference-costs](), [intelligence](), [models](), [nemotron](), [nvidia](), [open](), [operations](), [platform](), [results](), [retail](), [retail-and-consumer-packaged-goods](), [robotic](), [technology]() ### AI overview European retailers in the excerpt use open models and physical AI for commerce, product discovery, research and warehouse planning. Shopify reports that post-training Nemotron 3.5 Lightning for a Sidekick GraphQL subagent beat a leading frontier model on its internal accuracy benchmark while reducing serving costs 90% and making responses 7x faster; its vLLM optimization also improved five other use cases. Zooplus' AWS prototype reached 86% search accuracy. Photoroom standardized about 40,000 Vestiaire Collective listing images daily and reported higher sales liquidity and product views. For operations, Otto Group simulates warehouse changes before physical implementation, while ALLSIDES creates SimReady digital twins for store simulation and product tracking. L'Oréal uses accelerated molecular screening to narrow ingredient combinations before lab testing. These are company-reported results; the excerpt does not establish independent comparisons across deployments. ### Source excerpt European retailers are using open models and physical intelligence across use cases -- from a commerce platform cutting inference costs by 90% to to a fulfillment center where collaboration of different robotic fleets is simulated inside a digital twin. Frontier Results With Open Models Retail leaders and technology companies across Europe are using NVIDIA-accelerated open [...] ## The Price of Choice in Selective Prediction DevFeed: [The Price of Choice in Selective Prediction]() Original publisher: [Read original article]() Author: Daniel Cie\\'slak, Andrzej Czy\\.zewski Published: 2026-10-22T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [test](), [audit]() Tags: [audit](), [choice](), [classifiers](), [confidence](), [contracts](), [cost](), [coverage](), [cs-lg](), [error](), [experiments](), [gamma](), [gap](), [gate](), [hidden](), [image](), [it](), [linear](), [lower-bound](), [minimax](), [multimodal](), [power]() ### AI overview Refuting an existential risk-coverage claim requires ruling out every rule in the predeclared menu. The intersection-union test (IUT) does so without a multiplicity correction, but its sample cost grows with menu richness: for K-interval menus, the minimax sample size scales as K(1−κ)/(κγ²), up to logarithmic factors, in the stated regime, and exact IUT attains this rate. The lower bound holds even when the score distribution is known; the upper bound extends to menus of VC dimension d. Audit procedures that refuse to reject whenever any rule appears feasible can incur linear K cost, though a square-root rate may be possible; unknown heterogeneous risks can make linear cost unavoidable. Experiments report positive exact-hiding windows across tested VLM confidence-gate families, with weaker transfer on a cleaner multi-annotator target. Frozen image and multimodal classifiers show boundary behavior consistent with the guarantee and strong power against practical certificates. ### Source excerpt arXiv:2610.29067v1 Announce Type: new Abstract: A selective classifier may be claimed to satisfy a risk-coverage contract if some rule from a predeclared menu (a threshold or a union of $K$ score intervals) attains coverage at least $\kappa$ and error at most $\alpha$. Refuting this existential claim requires ruling out every permitted rule. The intersection-union test (IUT) achieves validity without a multiplicity correction, but its sample cost can increase with menu richness. We characterize this cost and compare it with the optimal audit. For a fixed interior risk target $\alpha$ and gap $\gamma$, the minimax sample size for K-interval menus at coverage floor $\kappa$ is $K(1-\kappa)/(\kappa\gamma^2)$, up to logarithmic factors, in the stated small-gap, large-menu regime; exact IUT attains this rate. The lower bound holds even with a known score distribution, while the upper bound extends to menus of VC dimension $d$. Procedures that refuse to reject whenever any rule appears feasible on the audit sample can pay a linear $K$ cost even when a square-root rate is possible. Reference-free testing reaches a smaller scale under constant-risk alternatives, whereas unknown heterogeneous risks can make the linear cost unavoidable. In simulations with $K=4$, the raw menu cost varies 7.7-fold between $\kappa=.5$ and $\kappa=.95$; after theorem-motivated normalization the endpoint values differ by approximately 5%. Held-out experiments with three VLM judges show positive exact-hiding windows in all 60 split-specific confidence-gate families, with 28/30 contracts chosen on audit-A remaining hidden on untouched audit-B. On a cleaner multi-annotator target, hiding is weaker and transfers in 3/10 resplits. Frozen image and multimodal classifiers show boundary behavior consistent with the IUT guarantee and strong power relative to practical certificates. ## Do Coding Agents Understand Why Systems Crash---and Why They Don't? DevFeed: [Do Coding Agents Understand Why Systems Crash---and Why They Don't?]() Original publisher: [Read original article]() Author: Chenxi Huang, Haonan Wang, Jinghe Zhang, Aleksandr Nogikh, Franjo Ivan\\v{c}i\\'c, Petros Maniatis, Eugene Wu, Kostis Kaffes, Junfeng Yang, Baishakhi Ray Published: 2026-10-22T04:00:00Z Content type: article Language: en Sources: [arXiv: Software Engineering]() Topics: [agentic-coding](), [AI-assisted coding](), [Linux Kernel](), [Fuzzing/Fuzz testing](), [code analysis](), [Artificial Intelligence]() Tags: [coding-agents](), [crash](), [cs-se](), [dataset](), [evaluation](), [execution](), [source-code](), [systems]() ### AI overview The study finds that current coding agents cannot reliably predict crashes in complex systems code. In tests using four agents and 934 Linux and OSS-Fuzz source-tree pairs that differ by one commit, agents tended to predict no crash, missing many real crashes; even the best missed more than a quarter. Agents also often misjudged whether relevant code would cause a crash, especially when the failure occurred elsewhere. Testing only crashing versions can overstate reasoning ability, since agents may flag suspicious code in both versions. The authors use paired source interventions to distinguish genuine crash reasoning from that behavior and argue paired evaluation is necessary. ### Source excerpt arXiv:2610.29332v1 Announce Type: new Abstract: Coding agents are increasingly capable of fixing real-world bugs, yet they still struggle with large systems codebases such as the Linux kernel. To understand why, we test how agents reason about execution in these codebases through crash reasoning: predicting from source code whether a given input crashes the system. A correct prediction alone, however, does not establish this ability, since it may come from merely spotting suspicious code. In this paper, we propose source interventions: we pair source trees that differ by a single commit, so that the same input crashes only one of them, and test whether agents predict crashes more often on the crashing tree. Experiments with four coding agents on SysRB, our dataset of 934 such pairs from Linux and OSS-Fuzz projects, yield three main findings. First, agents are biased toward predicting no crash: they rarely raise false alarms but miss many real crashes, and even the best agent misses over a quarter of them. Second, locating the code that determines the crash outcome is not enough: agents often misjudge its effect, especially when the crash occurs far from that code. Third, testing only on crashing code is misleading: agents that merely notice suspicious code predict crashes on both source trees, which would look like correct reasoning on the crashing tree alone. Together, these findings indicate that current coding agents cannot yet reliably reason about execution in complex systems codebases, and that paired evaluation is necessary to measure this ability. ## The End of Consensus Research? DevFeed: [The End of Consensus Research?]() Original publisher: [Read original article]() Author: Cl\\'ement Burgelin, Rachid Guerraoui, Zicong Hong, Yinan Hu, Anne-Marie Kermarrec, Rafael Pires, Rasmus Moorits Veski, Martijn de Vos Published: 2026-10-22T04:00:00Z Content type: article Language: en Sources: [arXiv: Distributed Computing]() Topics: [consensus algorithm](), [Agent Harness](), [consistency](), [Multi Agent Systems](), [vulnerability-research]() Tags: [agent](), [agent-harness](), [ai](), [autonomous](), [consensus](), [cs-ai](), [cs-dc](), [discovery](), [distributed-computing](), [experimentation](), [framework](), [implementation](), [key-value-store](), [knowledge-base](), [latency](), [leader](), [lean](), [lease](), [local](), [low-latency](), [modular](), [research]() ### AI overview AI helped discover A-greement, a consensus protocol for geo-replicated state machines, but the authors argue that human experts remain essential to consensus research. Across six deployments with 5 or 9 replicas, it reduced mean latency over the fastest baseline by 12.2% on write-heavy workloads and 33.3% on read-heavy workloads, while matching the baseline on read-only workloads. Its paths combine lease-authorized local reads, direct client learning for leader-ordered writes, and fast fixed-quorum per-key writes, chosen according to network topology and workload. The AI-greement harness brings together a protocol knowledge base, modular implementation skeleton, and Lean-based verification framework; its formalization also exposed previously unreported vulnerabilities in existing protocols. ### Source excerpt arXiv:2610.30114v1 Announce Type: new Abstract: Consensus is a fundamental problem in distributed computing and at the core of state machine replication (SMR). Despite decades of active human research, discovering practical low-latency solutions to the problem is still open, as it requires combining protocol knowledge, correctness reasoning, implementation, and experimental evaluation. We report on how AI helped us discover A-greement, a new consensus protocol for geo-replicated state machines that significantly outperforms state-of-the-art solutions. Averaged across six deployments with 5 or 9 replicas, A-greement reduces the mean latency of a geo-distributed key-value store over the fastest baseline by 12.2% and 33.3% on write-heavy and read-heavy workloads, respectively, and matches the fastest baseline on read-only workloads. A-greement combines lease-authorized local reads, direct client learning for leader-ordered writes, and a fast fixed-quorum per-key write path, selected according to network topology and workload. At the heart of our discovery lies AI-greement, an agent harness for the "autonomous" discovery, implementation, experimentation, and verification of candidate protocols. AI-greement integrates (1) a concept-oriented protocol knowledge base, (2) a modular implementation skeleton, and (3) a Lean-based verification framework with reusable safety definitions, reference specifications, and proofs. AI-greement is of independent interest: its Lean formalization uncovers several previously unreported vulnerabilities in existing consensus protocols, and it provides guidelines on how domain-specific harnesses can turn general-purpose coding agents into effective tools for (distributed) computing research. Yet we argue that human consensus experts remain essential before, during, and after the search: agents reshape the researcher's role rather than end consensus research. ## ACHEM: A Real-Time Digital Twin Framework with Channel and Radio Emulation DevFeed: [ACHEM: A Real-Time Digital Twin Framework with Channel and Radio Emulation]() Original publisher: [Read original article]() Author: Anil Gurses, Mihail L. Sichitiu Published: 2026-10-22T04:00:00Z Content type: article Language: en Sources: [arXiv: Networking]() Topics: [5G]() Tags: [4g](), [5g](), [antenna](), [channel](), [communication](), [cs-ni](), [deployment](), [digital-twin](), [eess-sp](), [emulation](), [fpgas](), [framework](), [i-q](), [mimo](), [mobility](), [models](), [multiple](), [node](), [open-source](), [optimizing](), [radio](), [real-time]() ### AI overview ACHEM is a software-based, end-to-end channel emulation framework for USRP-based wireless systems. It emulates systems at the I/Q level in a digital environment, avoiding specialized hardware such as USRPs, FPGAs, GPUs, or vehicles. It supports multiple transmitters and receivers, MIMO, multiple frequencies, differing sampling rates, real-time mobility through vehicle emulation, antenna radiation patterns, and multiple channel models. The authors report validation with GNU Radio, srsRAN 4G/5G, and OpenAirInterface. The abstract positions ACHEM as an alternative to costly, cumbersome testbeds and simulations that may simplify channel/radio behavior or omit full-stack production code. ### Source excerpt arXiv:2604.04742v2 Announce Type: replace-cross Abstract: Digital twins are becoming an important tool for designing, developing, testing, and optimizing next-generation wireless communication systems. Over the past decade, system softwarization has become a reality, and wireless communication systems are no exception. Software-Defined Radios (SDRs), in general, and Universal Software Radio Peripherals (USRPs), in particular, are often used for prototyping and testing advanced wireless systems. Unfortunately, there is currently no end-to-end, software-based, general-purpose testing environment for SDR-based systems: developers often rely on benchtop setups or even small testbeds, but those are costly and cumbersome to build. At the other end of the spectrum, simulations often rely on simplified channel/radio models and typically do not execute full-stack production code, which can increase development effort and reduce fidelity. In this paper, we propose ACHEM (A Channel Emulator), the first software-based, end-to-end wireless channel emulation environment and toolset for communication systems based on SDRs, specifically USRPs. With the proposed emulator and toolkit, any USRP-based system can be fully emulated at the I/Q level in a pure digital environment without requiring specialized hardware (e.g., vehicles, USRPs, FPGAs, or GPUs). The proposed emulator supports multiple transmitters and receivers, MIMO communications, multiple frequencies, heterogeneous sampling rates, real-time node mobility through vehicle emulation, antenna radiation patterns, and various channel models. ACHEM facilitates wireless digital twin development and deployment. ACHEM is validated with several popular open-source USRP-based wireless communication applications, including GNU Radio, srsRAN 4G/5G, and OpenAirInterface. ## Combining Long-Term Memory and LLMs Comes at a High Cost: From Prompt Injection to Persistent Compromise DevFeed: [Combining Long-Term Memory and LLMs Comes at a High Cost: From Prompt Injection to Persistent Compromise]() Original publisher: [Read original article]() Author: Pedro H. Barcha Correia, Matheus Azevedo de S\\'a, Paulo H. Gomes de Senna, Giovanni Mandel Martignago, Victor Takashi Hayashi, Charles Christian Miers, Marcos A. Simplicio Jr Published: 2026-10-22T04:00:00Z Content type: article Language: en Sources: [arXiv: Cryptography and Security](), [arXiv: Machine Learning]() Topics: [LLM security](), [ai security](), [prompt injection](), [AI Chat](), [Generative AI]() Tags: [attack](), [attacks](), [behavior](), [bypass](), [chatbots](), [chatgpt](), [claude](), [claude-sonnet](), [content](), [context](), [conversational-interfaces](), [conversations](), [cross](), [cs-ai](), [cs-cl](), [cs-cr](), [cs-lg](), [cybersecurity](), [datasets](), [defense](), [disinformation](), [llms](), [memory](), [prompt-injection]() ### AI overview The study describes memory injection as a way for malicious instructions to persist across conversations through long-term memory features. Its general threat model finds that all NIST attack goals are collectively achievable across ChatGPT Instant 5.3, Claude Sonnet 4.6, and Gemini 3 Flash, using vectors such as personal memories, projects, skills, organization instructions, and MCP-connected resources. The evaluated services could be made to persistently spread disinformation, generate harmful content, or simulate unresponsiveness in fresh sessions. The abstract reports that Claude's shared memory could extend compromise across an organization and expose user prompts. The authors derive defenses from the threat model and cybersecurity principles, and report that nearly half remain unimplemented across the services. These are abstract-level findings; the supplied text does not detail individual defenses or attack procedures. ### Source excerpt arXiv:2610.28810v1 Announce Type: new Abstract: The widespread adoption of generative artificial intelligence in recent years has brought a parallel rise in prompt injection and jailbreak attacks. By embedding malicious instructions, these exploits manipulate language models to bypass intended behavior and operational boundaries. In traditional chatbots, such attacks are confined to a single session because state memory is not preserved across conversations. However, persistent memory features in modern conversational interfaces enable cross-session attack propagation, establishing a threat vector known as memory injection. This work explores this emerging vulnerability in leading LLM-as-a-service platforms, highlighting its real-world severity and the need for countermeasures. While existing literature focuses on violations tailored to specific platform features, we present a general threat model. We demonstrate that all NIST attack goals are collectively achievable across ChatGPT Instant 5.3, Claude Sonnet 4.6, and Gemini 3 Flash under this model. We also show that memory injections can exploit multiple attack vectors, including personal memories, projects, skills, organization instructions, and Model Context Protocol (MCP)-connected resources. Our findings reveal that all evaluated services can be manipulated to persistently spread disinformation, generate harmful content, or simulate offline unresponsiveness, even in fresh sessions. In the particular case of Claude, shared-memory features allow an attacker to compromise multiple users within an organization and also exfiltrate user prompts. To address these vulnerabilities, we derive defense mechanisms from our proposed threat model and foundational cybersecurity principles. Our analysis indicates that nearly half of these mitigations remain unimplemented across the target services. Datasets and attack artifacts are available at https://github.com/PedroBarcha/Memory-injection. ## Efficient Path Query Processing in Relational Database Systems DevFeed: [Efficient Path Query Processing in Relational Database Systems]() Original publisher: [Read original article]() Author: Diego Rivera Correa, Mirek Riedewald Published: 2026-10-22T04:00:00Z Content type: article Language: en Sources: [arXiv: Databases]() Topics: [graph-database](), [rdbms](), [Graphs](), [SQL](), [DuckDB]() Tags: [constraint](), [cs-db](), [database](), [graph](), [processing](), [query](), [query-processing](), [regular-expressions](), [systems]() ### AI overview ReCAP helps relational database optimizers evaluate property-graph path queries by making early filtering opportunities explicit, so intermediate results with no possible path completion can be discarded sooner. It provides an abstraction for implementing early filtering for eligible property constraints: implement an NFA-style state transition function and a handful of functions modeled on user-defined aggregates. The abstract reports that DuckDB with ReCAP pushed constraints deeper into query plans and outperformed the state-of-the-art graph and relational DBMS compared by up to 400,000x across varied queries and input graphs; this is a reported maximum, not a general guarantee. ### Source excerpt arXiv:2604.02553v2 Announce Type: replace Abstract: Path queries are crucial for property graphs, and there is growing interest in queries that combine regular expressions over labels with constraints on property values of vertices and edges. Efficient evaluation of such general path queries requires that intermediate results be eliminated early when there is no possible completion to a full result path. Neither state-of-the-art (SOA) graph DBMS nor relational DBMS currently can do this effectively for a large class of queries. We show that this problem can be addressed by giving a relational optimizer ``a little help'' by specifying early filtering opportunities explicitly in the query. To this end, we propose ReCAP, an abstraction that greatly simplifies the implementation of early filtering techniques for any type of property constraint for which such early filtering can be derived. No matter how complex the constraint, one only needs to implement (1) an NFA-style state transition function and (2) a handful of functions that mirror those needed for user-defined aggregates. We show that when using ReCAP, a standard relational DBMS like DuckDB can effectively push property constraints deep into the query plan, beating the SOA graph and relational DBMS by a factor up to 400,000 over a variety of queries and input graphs. ## NVIDIA Nsight Python Tutorial: How to Profile GPU Kernels DevFeed: [NVIDIA Nsight Python Tutorial: How to Profile GPU Kernels]() Original publisher: [Read original article]() Author: Sana Hassan Published: 2026-10-22T03:59:12Z Content type: tutorial Language: en Sources: [MarkTechPost]() Topics: [Tutorial](), [Python](), [pandas]() Tags: [ai-infrastructure](), [artificial-intelligence](), [editors-pick](), [for-devs](), [gpu](), [kernels](), [nvidia](), [profile](), [python](), [software-engineering](), [staff](), [tech-news](), [technology](), [tutorial](), [tutorials]() ### AI overview Nsight Python profiles annotated GPU work from Python and returns hardware metrics in a pandas DataFrame. The tutorial shows setup on a Colab T4, including installing Nsight Compute 2026.2.1 or later and importing nsight before CUDA initializes. It explains that regions with multiple kernels require combine_kernel_metrics; repeated runs provide spread and confidence intervals, while StableMeasurement means relative standard deviation below 2%. Cache flushing and unlocked clocks are Nsight Python defaults, unlike plain ncu's base clock setting, and the ComputeClock field is a rated device attribute rather than the clock measured during execution. Examples use counters to assess compute versus memory limits, bytes to compare fused and unfused work, and custom collectors for per-run clocks and temperature. The experimental CUPTI backend profiles with lower overhead, but records duration only, without replay, cache flushing, or clock control. The tutorial also cautions that normalize_against changes AvgValue while leaving StdDev and Unit tied to the raw measurement. ### Source excerpt Discover NVIDIA Nsight Python, an open-source interface that allows developers to profile GPU kernels, analyze hardware metrics, evaluate cache and clock controls, and measure kernel fusion directly from Python code. The post NVIDIA Nsight Python Tutorial: How to Profile GPU Kernels appeared first on MarkTechPost. [Next page]()