# Lab Published articles for Lab. This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles. ## Bridging the Digital Divide: 25 Years of Empowerment in Panama DevFeed: [Bridging the Digital Divide: 25 Years of Empowerment in Panama]() Original publisher: [Read original article]() Author: Rebeca De La Vega Published: 2026-11-05T15:00:27Z Content type: article Language: en Sources: [Cisco Blogs]() Topics: [Learning](), [cisco networking](), [digital]() Tags: [anniversary](), [ccna](), [certification](), [cisco-networking-academy](), [digital](), [digital-divide](), [digital-skills](), [education](), [engineering](), [equity](), [future](), [graduation](), [hackathons](), [institutions](), [lab](), [learn-with-cisco](), [networks](), [packet](), [partnership](), [requirements](), [standard]() ### AI overview Cisco Networking Academy in Panama grew from its first academy in 2000 into a nationwide technical education program. Packet Tracer gave students simulated lab experience, and a partnership with the Technological University of Panama made CCNA certification a graduation requirement for Systems Engineering students. The academy also expanded through partnerships with INADEH, ITSE, Universidad Latina de Panama, and Universidad Tecnologica Oteima. Community networks in indigenous areas and national hackathons extended digital skills training. The source reports more than 50,000 skilled, certified students, alongside updated curriculum and participation in regional Cyber Games. ### Source excerpt Cisco Networking Academy celebrates its 25-year presence in Panama. From partnering with indigenous communities to setting new educational standards for technical education, we're bridging the country's digital divide one certification at a time. ## Workflow lab: Test ideas with code layers and the Figma agent DevFeed: [Workflow lab: Test ideas with code layers and the Figma agent]() Original publisher: [Read original article]() Author: Mari Kong Published: 2026-11-04T22:12:00Z Content type: tutorial Language: en Sources: [Figma Blog]() Topics: [Figma agent](), [Figma](), [User interface design](), [Code review]() Tags: [3](), [agent](), [alternatives](), [api](), [app](), [canvas](), [code](), [compare](), [designer](), [explore](), [feature](), [figma](), [figma-agent](), [ideas](), [lab](), [layers](), [new-feature](), [prototypes](), [teams](), [workflow]() ### AI overview Figma code layers let teams bring an existing app onto the canvas, generate interactive alternatives with the Figma agent, compare them side by side, and combine work. In the Pickup example, a designer clones a connected GitHub repository, asks the agent for three delivery-address options, then duplicates and manually edits a version. They can extract the prototype as editable design layers, record its screens as frames, and send design changes back to the code layer. The designer then combines the delivery flow with a teammate's map-grid exploration and asks the agent to describe changes and flag technical implications for developer review. The resulting prototype and spec support review before implementation and release; the article notes code layers are in open beta. ### Source excerpt Bringing code onto the canvas gives teams a shared space to compare working prototypes side by side, refine a direction, and collaborate on ideas with teammates. This workflow shows how a designer uses code layers and the Figma agent to explore a new feature. ## Does reflective LLM feedback survive a wet-lab-scale label budget? DevFeed: [Does reflective LLM feedback survive a wet-lab-scale label budget?]() Original publisher: [Read original article]() Author: Haoxuan Zeng, Xin Luo Published: 2026-11-03T05:00:00Z Content type: article Language: en Sources: [arXiv: Artificial Intelligence]() Topics: [Large Language Model](), [Data Mining & Modeling]() Tags: [acquisition](), [budget](), [correlation](), [cs-ai](), [feedback](), [lab](), [llm](), [model](), [random](), [rank](), [recovery]() ### AI overview Across 36 ProteinGym assays with 96 initial labels and six rounds of 96-label acquisition, full LLM control showed no detected test-rank-correlation advantage over random edits (Δz = −0.008; 95% CI −0.034 to +0.018), and equivalence was not established. Exploratory ablations found gains from LLM-proposed model edits: Fisher-transformed test Spearman rose by 0.043 with random acquisition, and normalized top-variant recovery rose by 0.023 with upper-confidence-bound acquisition. LLM acquisition alone did not detectably beat fixed adaptive rules, and repeated feedback did not detectably beat one initial recommendation. The results support constrained model editing in these experiments; added value from recursive reflection at this label budget remains unestablished. ### Source excerpt arXiv:2611.01908v1 Announce Type: new Abstract: Can reflective language-model feedback improve a biological predictor when each cycle buys only one plate of labels? We separate three possible sources of benefit: proposing model edits, choosing measurements, and revising decisions through repeated feedback. We replay $36$ ProteinGym assays, starting with $96$ labels and acquiring $96$ per round for six rounds, with a sealed test set and a shared edit schema. The pre-specified comparison of full LLM control against random edits detects no advantage in test rank correlation ($\Delta z=-0.008$, $95\%$ CI $[-0.034,+0.018]$); equivalence is also unestablished. Exploratory ablations nevertheless identify useful model editing. With random acquisition fixed, LLM edits improve Fisher-transformed test Spearman by $0.043$ over no edits; with acquisition fixed to upper confidence bound, they improve normalized top-variant recovery by $0.023$. LLM acquisition alone does not detectably outperform fixed adaptive rules, and repeated feedback does not detectably outperform one initial recommendation under full control. Thus, the observed gains support constrained specialist editing, while the incremental value of recursive reflection remains unestablished at this label budget. ## netlab: Testing Network Protocol Implementations DevFeed: [netlab: Testing Network Protocol Implementations]() Original publisher: [Read original article]() Published: 2026-11-02T07:18:00Z Content type: tutorial Language: en Sources: [ipSpace]() Topics: [Protocol (disambiguation)](), [BGP](), [developer setup]() Tags: [bgp](), [cisco](), [configuration](), [container](), [containers](), [eigrp](), [implementation](), [infrastructure](), [interoperability](), [ios](), [ios-xe](), [lab](), [labs](), [netlab](), [network-device](), [repository](), [server](), [source-code](), [test](), [testing]() ### AI overview The excerpt reports that netlab worked for testing OpenEIGRP-Cisco IOS interoperability and gives setup guidance for similar protocol tests. It recommends container labs via the `clab` provider for faster startup than virtual machines; IOL is suggested for IOS/XE, with IOLL2 faster but not guaranteed to support all features. For Nexus OS, it suggests obtaining the latest registered download and using vrnetlab to make a container. Labs can use one topology per directory or multiple topology files, with filenames specified to `netlab create` or `netlab up` in the latter case; defaults can be project-wide. The author recommends putting FRR settings in defaults and using wildcard groups to assign device types. Netlab's EIGRP support is described as minimal and potentially insufficient, while it can handle other configuration such as addressing, OSPF, BGP, and static routes. Custom Jinja2 templates are suggested for EIGRP configuration, with possible later work toward generic support or OpenEIGRP integration. ### Source excerpt Donnie Savage decided to use netlab to test OpenEIGRP-Cisco IOS interoperability (TL&DR: it works and the source code is in Donnie's EIGRP repository). He asked me for a few hints before setting up his test server. This is what I sent him; you might find it useful if you ever decide to use netlab to test a new network device/protocol implementation. Read more ... ## Anthropic expected to receive high 'junk' rating from credit agencies following IPO, FT report says DevFeed: [Anthropic expected to receive high 'junk' rating from credit agencies following IPO, FT report says]() Original publisher: [Read original article]() Author: Lucas Ropek Published: 2026-10-30T21:52:03Z Content type: news Language: en Sources: [TechCrunch | AI News & Artificial Intelligence]() Topics: [Finance]() Tags: [agencies](), [ai](), [ai-industry](), [anthropic](), [blackstone](), [chips](), [claude](), [compute](), [deal](), [finance](), [forecast](), [funding](), [in-brief](), [investment](), [investors](), [ipo](), [lab](), [range](), [rating](), [report](), [spending]() ### AI overview Blackstone reportedly now expects Anthropic to receive a double-B rating--the highest tier of speculative, or "junk," grade--after its IPO, reversing its earlier expectation of an investment-grade rating before 2027. This is a forecast, not a rating decision; Anthropic and Blackstone declined to comment. A high junk rating could make borrowing more difficult and costly as Anthropic continues its large spending on computing capacity, which has concerned investors. ### Source excerpt Blackstone's forecast is just a prediction, but naturally, a high junk rating would make it more difficult and costlier for Anthropic to raise the money to keep funding its massive spending that has undergirded its business. ## NetClaw on Heavy Networking: An Open Source Network Assistant DevFeed: [NetClaw on Heavy Networking: An Open Source Network Assistant]() Original publisher: [Read original article]() Author: John Capobianco Published: 2026-10-30T12:00:00Z Content type: article Language: en Sources: [Automate Your Network]() Topics: [AI Networking](), [Open Source](), [google-adk]() Tags: [actions](), [agent](), [ai](), [article](), [cisco](), [community](), [draft](), [engineer](), [engineering](), [environment](), [experiment](), [feedback](), [governance](), [lab](), [layers](), [maintainers](), [mcp](), [netclaw](), [network](), [network-automation](), [networking](), [open-source](), [protection]() ### AI overview NetClaw's excerpted discussion emphasizes trust through explicit approval: actions beyond reading network state require approval, with operational rules informed by engineering mistakes and outages. It describes Cisco DefenseClaw and NVIDIA OpenShell as additional protections to configure and evaluate in each environment, and recommends starting in a lab with read-only tasks while learning the tools' boundaries. NetClaw federation, called NCFED, is an Internet-Draft open for development and review. The project also keeps user stories, requirements, plans, and tests alongside implementation to help contributors understand the work. Maintainers, contributors, and engineers testing it in their labs are shaping the project. ### Source excerpt Back with Ethan Banks and Drew Conry-Murray on Packet Pushers to talk NetClaw: MCP, explicit approvals, layered protection, federation, and the community building it. ## How the UK Amazon Quick User Group grew in 2026 DevFeed: [How the UK Amazon Quick User Group grew in 2026]() Original publisher: [Read original article]() Author: Priyanka Bettaswamygowda Published: 2026-10-29T20:16:40Z Content type: article Language: en Sources: [Artificial Intelligence]() Topics: [AI for customer service automation](), [AI Chat]() Tags: [2026](), [actions](), [adoption](), [advanced](), [agent](), [agentic-ai](), [agentic-workflows](), [agents](), [ai](), [ai-assistants](), [amazon](), [amazon-quick](), [amazon-quick-suite](), [customer](), [customer-stories](), [data](), [faster](), [growth](), [intermediate-200](), [lab](), [uk](), [uncategorized](), [using]() ### AI overview The UK Amazon Quick User Group held three quarterly meetups in 2026 with more than 150 attendees, combining customer talks, product updates, networking and 75-minute hands-on workshops in sandbox accounts. Workshops covered custom chat agents, Quick Desktop with local files and an Outlook connector, Quick Apps, Flows, Automate and Research. Customer examples included Arriva's role-based mobile dashboards and scheduled reports for managers; Carsa's governed metric layer, which cut dashboards by 50% and recurring tasks from 45 minutes to 5; and Propulse Lab's multi-agent pipeline, which reduced ticket handling from 7 minutes to 70 seconds and automated 80% of routine workflows. The organizers adjusted the format based on feedback, adding quizzes, a specialist panel and networking time; attendees increasingly moved from listening to contributing and presenting. ### Source excerpt The UK Amazon Quick User Group's 2026 roundup: three quarterly meetups, 150+ attendees, and customer stories from Arriva, Carsa, and Propulse Lab on using agentic AI and Amazon Quick to turn data into faster decisions, higher satisfaction, and scalable growth. ## How Zurich is shaping the future of algorithms, 70 years in DevFeed: [How Zurich is shaping the future of algorithms, 70 years in]() Original publisher: [Read original article]() Author: Mariana Rajado Silva Published: 2026-10-29T14:15:00Z Content type: news Language: en Sources: [IBM Research]() Topics: [Algorithms](), [Cryptography](), [AI Models](), [quantum mechanics]() Tags: [agentic](), [agentic-ai](), [ai](), [algorithms](), [applications](), [computer-science](), [computing](), [cpus](), [cryptography](), [digital-twins](), [future](), [ibm](), [lab](), [q-a](), [vision]() ### AI overview Alessandro Curioni says IBM Research is developing foundational algorithms that combine classical computing, quantum computing and AI across five areas: differential equations and dynamical systems; linear algebra and Hamiltonian problems; optimization and combinatorial problems; probability, statistics and stochastic processes; and cryptography. He envisions CPUs, GPUs and QPUs working together under an agentic AI orchestration layer, enabling new approaches to complex problems and more accurate digital twins for science and business. He also credits IBM Research's ability to attract and retain talent, its global reach, and its connection between research and industry as key strengths. The excerpt presents his views and priorities, not specific algorithm results or implementation requirements. ### Source excerpt As IBM's Zurich Lab turns 70, director Alessandro Curioni speaks about the celebrations, his work on Algorithms & Applications, and his vision for the future. ## 7 Tools My Home Lab Depends On in Late 2026 DevFeed: [7 Tools My Home Lab Depends On in Late 2026]() Original publisher: [Read original article]() Author: Brandon Lee Published: 2026-10-29T12:17:28Z Content type: article Language: en Sources: [Virtualization Howto]() Topics: [Homelab](), [Tool](), [AI Chat](), [Self Hosted AI Tools]() Tags: [2](), [2026](), [3](), [about](), [added](), [agent](), [agent-skills](), [but](), [first](), [home](), [home-lab](), [home-server](), [howto](), [i](), [lab](), [my](), [post](), [thinking](), [this](), [tools](), [virtualization]() ### AI overview The author's late-2026 home lab setup combines tools for container and Kubernetes management, agent access, SSH, backups, and documentation. Komodo handles Docker updates and GitOps workflows; Aptakube aggregates Kubernetes resources and integrates with ArgoCD, though it is commercial. Termix provides browser-based SSH and other modular tools, including RDP. Veeam Kasten backs up Kubernetes workloads to local storage and a Synology NAS, and NetBox serves as the intended lab source of truth, populated with help from Hermes Agent. The author uses Hermes for troubleshooting, infrastructure reports, and inventory work, powered by a $20/month Codex subscription because local hardware and security validation for open-weight models are lacking. Octelium gates agent access to services and APIs with policies such as allowing GET while blocking POST; the author is still testing it and recommends network segmentation as an initial control. ### Source excerpt I have covered a LOT of ground in the home lab this year so far and we aren't even completely through the year as of yet. But, I was thinking... The post 7 Tools My Home Lab Depends On in Late 2026 appeared first on Virtualization Howto. ## Block Unveils Dessa, a Full-Duplex Speech Model Trained Only on Speech DevFeed: [Block Unveils Dessa, a Full-Duplex Speech Model Trained Only on Speech]() Original publisher: [Read original article]() Author: Michal Sutter Published: 2026-10-29T07:50:39Z Content type: article Language: en Sources: [MarkTechPost]() Topics: [Speech Processing](), [AI Models](), [Local AI](), [Inference](), [Large Language Model](), [asr]() Tags: [agents](), [ai](), [ai-research](), [api](), [architecture](), [artificial-intelligence](), [audio](), [audio-language-model](), [benchmark](), [benchmarks](), [blackwell](), [block](), [compute](), [cpus](), [dessa](), [editors-pick](), [full-duplex-speech-model](), [lab](), [language-model](), [macbook](), [macbook-pro](), [model](), [speech](), [speech-to-speech](), [staff](), [technology](), [voice-ai]() ### AI overview Block's research preview Dessa is a full-duplex speech-to-speech model trained only on human conversation audio, with no text or pretrained models, according to the lab. It listens while speaking and is shown in demos using backchannels, interruptions, laughter, and changes in volume. Block says it runs locally faster than real time on a MacBook Pro M4 Max or better and was trained on one NVIDIA GB200 NVL72 rack. The excerpt contrasts this with Moshi's text-model backbone, time-aligned text and audio tokens, and reported use of 1,016 H100 GPUs. Block has not published Dessa's architecture, benchmark scores, parameter count, weights, API, or price; access is described as waitlist-only. ### Source excerpt Block's AI lab released Dessa, a full-duplex speech-to-speech model trained only on human conversation audio. It runs locally on a MacBook Pro M4 Max, but Block has published no size, benchmarks or weights. The post Block Unveils Dessa, a Full-Duplex Speech Model Trained Only on Speech appeared first on MarkTechPost. ## OffSec Launches Free PHP2Shell Lab Following Critical Vulnerability Discovery DevFeed: [OffSec Launches Free PHP2Shell Lab Following Critical Vulnerability Discovery]() Original publisher: [Read original article]() Author: CIO Published: 2026-10-29T05:43:23Z Content type: news Language: en Sources: [CIO]() Topics: [Security research](), [cybersecurity and infrastructure security agency](), [Learning](), [exploit chaining](), [known exploitable vulnerabilities](), [Memory Safety]() Tags: [accelerated](), [ai](), [ai-assisted](), [amazon](), [artificial-intelligence](), [artificial-intelligence-security](), [attack](), [attacks](), [cisa](), [cloud-security](), [crash](), [critical](), [critical-vulnerability](), [cve-2026-109618](), [discovery](), [edition-us](), [free](), [lab](), [language-en](), [launches](), [offsec](), [php2shell](), [security](), [storytype-press-release](), [vulnerability](), [vulnerability-discovery]() ### AI overview OffSec released a free, interactive PHP2Shell lab in Proving Grounds Play for security practitioners to study CVE-2026-109618 and CVE-2026-109619 in a controlled setting. The flaws can let unauthenticated attackers crash or take over servers through crafted requests; the document also describes possible memory disclosure. OffSec recommends patching PHP itself to the listed fixed versions, checking vendor advisories, rebuilding affected container images, and restarting PHP processes. Temporary measures include disabling session upload progress and, on PHP 8.5, fatal error backtraces; these reduce exposure but do not replace patching. The article reports active exploitation and notes a separate, unfixed issue that may affect fully patched servers with an unusual session setup, so it advises checking CISA and PHP for current guidance. ### Source excerpt New hands-on lab lets security practitioners explore PHP2Shell exploitation firsthand as AI-assisted research reshapes vulnerability discovery. OffSec today announced the release of a free, interactive hands-on lab in Proving Grounds Play (PG Play) to help security practitioners examine and defend against the critical "PHP2Shell" vulnerabilities (CVE-2026-109618 and CVE-2026-109619) firsthand. Developed in response to AI-accelerated vulnerability discovery and active exploits targeting core PHP engines in the wild, the lab provides penetration testers and enterprise defenders with a safe, controlled environment to study the decade-old memory corruption flaws and implement timely server mitigations. The two critical flaws sat inside PHP, the language behind roughly 70% of websites whose server-side language is known, for more than a decade. Then researchers at Wiz, a Google-owned cloud security company, pointed AI at the code. They found the first flaw Aug. 19. The next day they used it to take over a Drupal site without a password. The day after that, it worked on WordPress. PHP disclosed the flaws Oct. 21 as CVE-2026-109618 and CVE-2026-109619. An attacker with no login can send a series of crafted web requests to crash a server or take it over. PHP's severity scores rate both vulnerabilities as complex to exploit, even though working attacks already existed. The Cybersecurity and Infrastructure Security Agency (CISA) recommends patching immediately. Attackers are moving fast. Amazon calls PHP2Shell broadly exploited in the wild. By Oct.24, researchers had confirmed attackers running commands on vulnerable honeypot servers. CISA added both flaws to its Known Exploited Vulnerabilities catalog on Oct. 26. "PHP is one of those technologies attackers love to see," said Keegan Bowen, principal DevOps lead at OffSec. "It has been a great attack vector over the years with many exploits and potential abuses. PHP wasn't designed with security from the start." OffSec, which ## OffSec Launches Free PHP2Shell Lab Following Critical Vulnerability Discovery DevFeed: [OffSec Launches Free PHP2Shell Lab Following Critical Vulnerability Discovery]() Original publisher: [Read original article]() Author: CyberNewswire Published: 2026-10-28T20:45:42Z Content type: news Language: en Sources: [HackerNoon]() Topics: [Exploit](), [Security research](), [Learning](), [exploit chaining]() Tags: [ai](), [critical](), [cyber-attack](), [cyber-security-awareness](), [cyber-threats](), [cybernewswire](), [cybersecurity](), [cybersecurity-tips](), [exploitation](), [free](), [good-company](), [hands-on-lab](), [lab](), [offsec](), [php2shell](), [press-release](), [research](), [security](), [vulnerability](), [vulnerability-discovery]() ### AI overview PHP disclosed two critical PHP memory corruption flaws, CVE-2026-109618 and CVE-2026-109619, after Wiz researchers used AI-assisted work to develop exploits that reportedly enabled passwordless takeovers of Drupal and WordPress sites. Crafted requests can leave interrupted state in a PHP worker, potentially enabling code execution with web server privileges or exposing process memory. The flaws were reported exploited in the wild; CISA added them to its Known Exploited Vulnerabilities catalog and recommends immediate patching. OffSec released a free, controlled PHP2Shell lab in Proving Grounds Play for hands-on study. Its stated response steps are to update PHP to a fixed release, check vendor advisories, rebuild affected containers and restart PHP processes; temporarily disable session upload progress and, on PHP 8.5, fatal error backtraces; and investigate logs and rotate potentially exposed credentials. These temporary controls do not replace patching. The text also notes a separate, unpatched session-configuration flaw and says recommendations may change as research continues. ### Source excerpt New hands-on lab lets security practitioners explore PHP2Shell exploitation firsthand as AI-assisted research reshapes vulnerability discovery. ## Testing Agentic Service Systems: An Agentic Living Lab Methodology Across the Physical-Digital Continuum DevFeed: [Testing Agentic Service Systems: An Agentic Living Lab Methodology Across the Physical-Digital Continuum]() Original publisher: [Read original article]() Author: Zhongjie Wang, Xiang He, Xiao Wang, Mingyi Liu, Shuang Yu, Hanchuan Xu Published: 2026-10-28T04:00:00Z Content type: article Language: en Sources: [arXiv: Software Engineering]() Topics: [Agentic AI](), [Software Engineering](), [Access Control]() Tags: [abstract](), [admission](), [agent](), [agentic](), [agents](), [architecture](), [arxiv](), [authorization](), [based](), [business](), [but](), [change](), [changes](), [complete](), [cs-se](), [digital](), [lab](), [methodology](), [service](), [systems](), [testing]() ### AI overview Task success alone cannot establish that an agentic service system behaved acceptably: agents may act without authorization, leave obligations unresolved, or produce results that vary by execution setting. The proposed Agentic Living Lab methodology tests a service episode with its delegated roles, obligations, world state, and evidence; separates agents under test from environment actors; and makes differences across software, enterprise systems, simulations, and physical settings explicit through realization, grounding, and semantic-delta disclosures. Its architecture separates experiment control, participants, world realization, and evidence evaluation, while runtime semantics distinguish intent, normative admission, and factual reconciliation. Real stakeholder grounding, participant calibration or validation, and governed feedback are required for a Living Lab, though humans need not participate synchronously in every run. JSON Schemas, deterministic fixtures, and a limited reference runner check selected rules, but do not constitute a complete platform or empirical validation. Agent quality, proxy validity, and stability across substrates remain open empirical questions. ### Source excerpt arXiv:2610.35978v1 Announce Type: new Abstract: Task success is insufficient for testing agents that act for principals, change persistent business state, and affect physical processes: a useful result may be unauthorized, leave obligations unresolved, or depend on the execution substrate. We propose an Agentic Living Lab methodology organized around three questions: what is the test unit, who and what participate, and under what reality is the test executed? The unit is a service episode containing delegated roles, obligations, world state, and evidence. The method separates agents under test from environment actors and distinguishes real humans, human traces, rule surrogates, and agent-based proxies. Realization Profiles, capability-declared Reality Substrate bindings, Grounding Profiles, and Semantic Delta disclosures make changes across software, enterprise systems, simulation, and physical settings explicit. A four-plane reference architecture separates experiment control, participants, world realization, and evidence evaluation, with cross-cutting authority and governance controls. Runtime semantics distinguish intent, normative admission, and factual reconciliation. Real stakeholder grounding, participant calibration or validation, and governed feedback are required for the Living Lab claim, but not synchronous human participation in every run. Released JSON Schemas, deterministic fixtures, and a limited reference runner make selected rules machine-checkable; they are not a complete platform or empirical validation. Constructed travel and industrial designs illustrate authorization and fulfillment failures, proxy sensitivity, and substrate changes. Agent quality, proxy validity, and cross-substrate effect stability remain empirical questions. ## MapPFN: Learning Causal Perturbation Maps in Context DevFeed: [MapPFN: Learning Causal Perturbation Maps in Context]() Original publisher: [Read original article]() Author: Marvin Sextro, Weronika K{\\l}os, Gabriel Dernbach Published: 2026-10-28T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Machine learning](), [Fine-tuning](), [generalization in machine learning](), [Model Development](), [Inference]() Tags: [abstract](), [arxiv](), [code](), [context](), [cs-lg](), [data](), [datasets](), [decoupling](), [distribution](), [effect](), [effective](), [estimator](), [existing](), [experiments](), [fine](), [fine-tuning](), [https](), [inference](), [lab](), [learning](), [maps]() ### AI overview MapPFN is designed to adapt perturbation-effect predictions to new biological contexts by using in-context learning over sequences of experiments to estimate post-perturbation distributions. It is pre-trained as a prior-data fitted network on a synthetic biological prior with causal interventions, then can use new interventional evidence at inference time; the abstract also reports improved predictions across contexts when synthetic pre-training is combined with fine-tuning on real single-cell data. The source does not provide quantitative results or details of the synthetic prior. ### Source excerpt arXiv:2601.21092v4 Announce Type: replace Abstract: Planning effective interventions in biological systems requires treatment-effect models that adapt to unseen biological contexts by identifying their specific underlying mechanisms. Yet single-cell perturbation datasets span only a handful of biological contexts, and existing methods cannot leverage new interventional evidence at inference time to adapt beyond their training data. To meta-learn a perturbation effect estimator, we present MapPFN, a prior-data fitted network (PFN) pre-trained on a synthetic biological prior with causal interventions, decoupling pre-training from limited wet-lab data. Unlike existing methods, MapPFN uses in-context learning to map a sequence of experiments to a post-perturbation distribution, enabling a single pre-trained model to adapt to new datasets and gene sets. Our results show that combining synthetic pre-training with fine-tuning on real single-cell data improves perturbation predictions across biological contexts. The code, model and data are available at https://marvinsxtr.github.io/MapPFN. ## How to Test an iOS App Without an iPhone: Simulators vs. Real Device Clouds DevFeed: [How to Test an iOS App Without an iPhone: Simulators vs. Real Device Clouds]() Original publisher: [Read original article]() Author: Tejas Patel Published: 2026-10-23T13:26:13Z Content type: tutorial Language: en Sources: [TestGrid]() Topics: [iOS](), [iphone](), [Emulator](), [Test automation](), [mobile-app-performance](), [mobile devops]() Tags: [app](), [apple](), [capabilities](), [compatibility](), [developers](), [hardware](), [independent](), [ios](), [ios-mobile-app-testing](), [iphone](), [lab](), [mobile-app-testing](), [models](), [multiple](), [operating-system](), [performance](), [release](), [screen](), [small](), [system](), [test](), [vs]() ### AI overview You can test iOS apps without owning an iPhone using Xcode Simulator or a remote real-device cloud. Simulator runs on a Mac and is useful for early checks of navigation, layouts, and basic behavior, but it cannot fully reproduce hardware features, iPhone performance constraints, sensor interactions, or reliable end-to-end push notification delivery. A device cloud provides remote access to physical iPhones for manual or Appium/XCUITest automation, including hardware and network checks, without maintaining a device lab. To run an app on your own iPhone without a paid Apple Developer Program membership, you still need a Mac with Xcode, a free Apple Account, and the iPhone; Xcode can sign through a Personal Team. Free provisioning is limited to 10 App IDs and three devices, with registrations expiring after seven days and requiring rebuild and reinstall. The excerpt recommends combining simulator checks with real-device validation for broader coverage. ### Source excerpt Apple's iPhone lineup includes multiple models, each with different hardware capabilities, operating system versions, screen sizes, and performance characteristics. And testing across these combinations is important to catch compatibility issues before release. However, buying multiple iPhones for testing, keeping them updated, and maintaining a device lab can be expensive, particularly for independent developers and small [...] ## Arduino® App Lab 0.11 is here DevFeed: [Arduino® App Lab 0.11 is here]() Original publisher: [Read original article]() Author: Arduino Team Published: 2026-10-22T15:05:11Z Content type: release Language: en Sources: [Arduino Blog]() Topics: [Arduino](), [project release](), [VENTUNO Q](), [UNO Q](), [Learning]() Tags: [about](), [app](), [app-lab](), [arduino](), [arduino-app-lab](), [arduino-app-lab-0-11](), [arduino-lab](), [board](), [build](), [code](), [deploy](), [device](), [featured](), [getting](), [great](), [help](), [here](), [lab](), [means](), [production](), [project](), [release]() ### AI overview Arduino App Lab 0.11 adds immutable app releases with locked dependencies and precompiled sketches for consistent deployment to compatible boards; remote dependencies such as AI models download at install time. The in-editor Agent Panel uses the open app and files as context and can update them during chat, while Agent Mode remains a separate full-window, agent-centered workflow. A local Pose Estimation Brick for VENTUNO Q tracks 17 body points at up to 10 fps and recognizes standard poses without separate training; custom poses use positive and negative photo folders. Other additions include importing compatible LLM models from Hugging Face, streaming phone microphone and IMU data over WebSocket with optional password protection and encryption, and a TPS positioning brick using nearby Wi-Fi, cellular, and Bluetooth LE signals. ### Source excerpt You build something great on your Arduino® UNO™ Q board. Now what? Arduino App Lab 0.11 is about taking your project further, whether that means shipping it to another board, getting help while you code, or teaching your device to recognize a tennis swing. Here is what's new! App release: deploy to production, at scale [...] The post Arduino® App Lab 0.11 is here appeared first on Arduino Blog. ## Claude Designed 211 Proteins in Two Days. The Memes-in-Slack Phase Took Two Months DevFeed: [Claude Designed 211 Proteins in Two Days. The Memes-in-Slack Phase Took Two Months]() Original publisher: [Read original article]() Author: Penicillin Published: 2026-10-22T14:38:39Z Content type: article Language: en Sources: [HackerNoon]() Topics: [Claude](), [anthropic](), [Slack]() Tags: [about](), [account](), [adaptyv-bio](), [ai](), [ai-drug-discovery](), [ai-in-healthcare](), [all](), [alphafold](), [american-express](), [analysts](), [animals](), [anthropic](), [biotech](), [citrini-research](), [claude](), [claude-designed-proteins](), [days](), [design](), [lab](), [memes](), [model](), [protein-design](), [proteins](), [research](), [slack]() ### AI overview Citrini Research used an unreleased Claude model at Anthropic to design scorpion-toxin binders in 64 hours, but the work did not produce an antivenom. Twist Bioscience and Adaptyv Bio synthesized and tested 211 designs; 17 met binding criteria, including 11 of 60 single-toxin designs, while none of 151 multitoxin designs bound more than one toxin. No neutralization tests in cells or animals were run, and broader testing raised questions about selectivity. The excerpt describes the team waiting roughly two months for lab results and argues that synthesis and assay capacity, plus later clinical development, may now constrain progress as computational design speeds up. External experts characterize AI-designed molecules as complementary research tools that still require validation and standardized manufacturing. ### Source excerpt Citrini Research used an unreleased Anthropic model to design scorpion toxin binders in 64 hours, then spent two months waiting on a wet lab. 17 of 211 bound. ## Your First Proxmox Server Can Do More Than You Think DevFeed: [Your First Proxmox Server Can Do More Than You Think]() Original publisher: [Read original article]() Author: Brandon Lee Published: 2026-10-22T12:35:43Z Content type: article Language: en Sources: [Virtualization Howto]() Topics: [Homelab](), [Proxmox](), [systems](), [virtualization](), [virtual machines](), [Tutorial](), [Service fabric]() Tags: [ansible](), [automate](), [automation](), [backing-up](), [backup](), [backups](), [clone](), [cloud](), [cloud-init](), [cloud-native](), [cloud-technologies](), [cluster](), [clustering](), [clusters](), [configuration](), [hardware](), [home-lab](), [home-server](), [lab](), [proxmox](), [server](), [virtualization]() ### AI overview A single Proxmox host can support substantial lab practice, though virtual nodes on one machine do not provide physical host redundancy. The article recommends separating lab and self-hosted services with distinct VLANs and subnets, while keeping switching and firewalling outside the host to preserve network access if Proxmox fails. Nested Proxmox VMs can be used to practice clustering and shared storage. A dedicated VM is the recommended place to run Docker; Proxmox also offers native LXC containers and, from v9.1, OCI image support that runs through LXC. Snapshots provide rollback points, while cloning lets you test changes on a copy. VM templates, cloud-init and optionally Packer and Ansible enable repeatable guest setup. A small Kubernetes cluster can teach node and scheduling basics, but all nodes stop if the physical host goes down. VM backups to local storage can teach the process, though the article says this is not ideal for data resilience. ### Source excerpt Recently, I wrote up an article looking at the things that you can skip when you want to stand up a really great Proxmox lab, even with little hardware. You... The post Your First Proxmox Server Can Do More Than You Think appeared first on Virtualization Howto. ## ChemBART: Pre-training a Reaction Language Model for Comprehensive Synthesis Planning DevFeed: [ChemBART: Pre-training a Reaction Language Model for Comprehensive Synthesis Planning]() Original publisher: [Read original article]() Author: Kenan Li, Yijian Zhang, Jin Wang, Haipeng Gan, Zeying Sun, Xiaoguang Lei, Hao Dong Published: 2026-10-22T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Large Language Model](), [LLM Techniques](), [llm-reasoning]() Tags: [classification](), [cs-cl](), [cs-lg](), [generation](), [integration](), [lab](), [language-model](), [large-language-models-llms](), [monte-carlo-tree-search](), [optimize](), [planning](), [reaction](), [regression](), [reinforcement-learning](), [route](), [synthesis](), [training]() ### AI overview ChemBART uses BART pre-trained to fill in masked parts of complete reactant-reagent-product expressions, enabling precursor and reagent generation, temperature-yield regression, and molecular property classification. The authors integrate it with Monte Carlo tree search, using the model to optimize policy and value functions for multi-step route design. The abstract reports wet-lab validation of proposed routes and conditions, with shorter pathways and yields over 20% higher than established literature benchmarks. The supplied text does not give implementation details or specify the scope of that comparison. ### Source excerpt arXiv:2601.02915v2 Announce Type: replace Abstract: Recent advances in large language models (LLMs) have demonstrated transformative potential across diverse fields. However, in computer-aided synthesis planning (CASP), current LLM methodologies are often restricted to single-molecule representations, limiting their application to isolated, single-task scenarios such as precursor prediction. Addressing this gap, we present ChemBART, a BART-based language model that adopts a reaction-centric pre-training strategy. By leveraging a mask-filling pre-training task on holistic reaction expressions (Reactant > Reagent > Product), ChemBART effectively solves a variety of chemical problems, including precursor/reagent generation, temperature-yield regression, and molecular property classification. Building on these capabilities, we demonstrate the integration of ChemBART with Monte Carlo tree search (MCTS) for multi-step synthesis route design, where the model serves to optimize the policy and value functions within a reinforcement learning framework. Beyond single-molecule pre-trained LLMs that are constrained to specific applications, ChemBART unifies diverse chemical tasks to facilitate synthesis planning with comprehensive implementation information. Most significantly, the proposed multi-step synthesis routes and reaction conditions were validated in the wet lab, confirming the discovery of shorter synthetic pathways with a >20% yield improvement over established literature benchmarks. Our work highlights the potential of reaction-focused pre-training and showcases the broad utility of ChemBART in closing the loop between in silico prediction and wet-lab synthesis. ## Run SAM 2 on a Cloud GPU with Runpod (Step-by-Step Guide) DevFeed: [Run SAM 2 on a Cloud GPU with Runpod (Step-by-Step Guide)]() Original publisher: [Read original article]() Author: Shaamil Karim Published: 2026-10-22T02:42:59Z Content type: tutorial Language: en Sources: [Runpod Blog.]() Topics: [Learning](), [jupyter](), [AI Infrastructure](), [Cloud]() Tags: [benchmarks](), [conda](), [cuda](), [customization](), [datasets](), [dependencies](), [deploy](), [developers](), [documentation](), [git](), [git-clone](), [github](), [gpu](), [guide](), [how-to](), [jupyter](), [lab](), [meta](), [model](), [sam]() ### AI overview The excerpt describes running Meta's SAM 2 on a Runpod GPU through Jupyter Lab. It provides a repository clone command and dependency setup commands using a virtual environment, then describes an example where a prompted point guides an image mask over an object. It notes that Jupyter Lab on port 8888 may take a few minutes to connect, and recommends checking `nvidia-smi` and consulting documentation or community support if issues persist. SAM 2 is presented as supporting promptable segmentation of images and videos, including multiple objects, with customization possible through parameter changes and specialized training. Detailed GPU requirements and complete deployment steps are not included in the supplied excerpt. ### Source excerpt Learn how to deploy Meta's Segment Anything Model 2 (SAM 2) on a Runpod GPU using Jupyter Lab. This guide walks through installing dependencies. ## Deploy a Stable Diffusion UI on Runpod in Minutes DevFeed: [Deploy a Stable Diffusion UI on Runpod in Minutes]() Original publisher: [Read original article]() Author: Zhen Lu Published: 2026-10-22T02:42:59Z Content type: tutorial Language: en Sources: [Runpod Blog.]() Topics: [stable-diffusion](), [automatic1111](), [CI/CD]() Tags: [cloud-compute](), [cuda](), [deploy](), [diffusion](), [guide](), [guides](), [huggingface](), [installer](), [jupyter](), [lab](), [model](), [pod](), [pytorch](), [stable-diffusion](), [template](), [train](), [ubuntu](), [ui]() ### AI overview Runpod's Stable Diffusion template, created by TheLastBen, is launched as a Pod and configured through Jupyter Lab: select RNPD-A1111 to run the A1111 UI, or a DreamBooth notebook to train a model. The excerpt also names a ControlNet installer, offset_noise for DreamBooth, and HuggingFace integrations. It notes that linked user videos may demonstrate a different template. ### Source excerpt This post shows how to deploy a full Stable Diffusion UI using a community-created template on Runpod. With just a few clicks, you can generate. ## Year-old AI research lab Flapping Airplanes raises $1B at a reported $6B valuation DevFeed: [Year-old AI research lab Flapping Airplanes raises $1B at a reported $6B valuation]() Original publisher: [Read original article]() Author: Duncan Riley Published: 2026-10-21T22:18:30Z Content type: news Language: en Sources: [SiliconANGLE]() Topics: [AI Research](), [Artificial Intelligence](), [AI Development](), [Frontier AI]() Tags: [ai](), [ai-models](), [ai-research](), [ai-training](), [aidan-smith](), [algorithms](), [artificial-intelligence](), [asher-spector](), [atreides-management](), [ben-spector](), [bloomberg](), [board](), [chip](), [company](), [compute](), [computer-science](), [data-efficiency](), [flapping-airplanes](), [funding](), [gpus](), [gv](), [index-ventures](), [inference](), [jump-trading](), [kleiner-perkins](), [lab](), [lux-capital](), [mark-xu](), [megakernels](), [neuralink](), [news](), [nvidia](), [prod](), [raises](), [sequoia-capital](), [series-a](), [shardul-shah](), [startup](), [the-latest](), [thunderkittens](), [top-story-6](), [valor-equity-partners](), [valuation](), [xtx-ventures]() ### AI overview Flapping Airplanes raised $1 billion at a reported $6 billion valuation about a year after its founding, though it has no commercial product yet. The lab aims to build AI models that use far less data and spent much of its first year developing GPU primitives. It describes its megakernel framework as able to run a full training run, inference serving, and a video game within one GPU kernel, allowing memory-bound inference to overlap with compute-bound training. The company says this work could reshape AI's cost structure; these are company claims, not independently established results in the excerpt. Index Ventures and Lux Capital co-led the Series A. Investors have put $1.18 billion into the company, and Bloomberg reported the valuation had roughly quadrupled since the seed round. ### Source excerpt Artificial intelligence research lab Flapping Airplanes Inc. today announced it has raised $1 billion in new funding, about a year after it was founded. Investors valued the startup at $6 billion in the deal, Bloomberg reported. No commercial product has come out of the lab yet. Flapping Airplanes is trying to build AI models that [...] The post Year-old AI research lab Flapping Airplanes raises $1B at a reported $6B valuation appeared first on SiliconANGLE. ## Train in Simulation, Deploy on a Real Robot: Microduck with Isaac Lab 3.0 DevFeed: [Train in Simulation, Deploy on a Real Robot: Microduck with Isaac Lab 3.0]() Original publisher: [Read original article]() Author: Oluwaseun Doherty; Oyindamola Omotuyi; Richard; Asier Arranz Published: 2026-10-21T17:58:42Z Content type: tutorial Language: en Sources: [Hugging Face]() Topics: [Isaac](), [Robotics Simulation](), [Isaac Sim](), [Open Source](), [Robot Manipulation](), [Robot Navigation]() Tags: [arena](), [bipedal-robot](), [challenge](), [compatibility](), [competition](), [configuration](), [deploy](), [export](), [hugging-face](), [isaac](), [lab](), [newton](), [onnx](), [open-source](), [policy](), [robot](), [robotics](), [simulation](), [test](), [train]() ### AI overview The tutorial trains a Microduck walking policy in Isaac Lab 3.0 with Newton, exports it to ONNX, and deploys it through the robot's onboard runtime. It uses GPU-parallel simulation, Newton's MJWarp solver, Better Actuator Models, and randomized robot and environment conditions--including servo friction, delay, and backlash--to help the policy handle hardware variation. Training requires Linux with a supported NVIDIA GPU and driver, Git, and uv; deployment also requires a configured Microduck and working onboard software. The policy uses 61 observations and 14 joint-position actions, whose order, normalization, and scaling must match across training and deployment. Playback exports both TorchScript and ONNX, but the robot runtime uses ONNX. The excerpt gives commands for training, evaluation, export, transfer, and loading, and notes that Microduck Arena accepts compatible policies up to 64 MiB. ### Source excerpt *This post was written in collaboration with Antoine Pirrone, Augustin Crampette, Matthieu Lapeyre, Coralie Deplanne, Anne-Charlotte Passanisi, and Thomas Wolf.* Microduck is a newly released affordable open-source bipedal robot by Pollen Robotics and Hugging Face. And now you can train Microduck to walk with Isaac Lab and Newton. Test the policy, export it as ONNX, and deploy it on the real robot. ## No Need for Retirement When Utopia Is Around the Corner DevFeed: [No Need for Retirement When Utopia Is Around the Corner]() Original publisher: [Read original article]() Author: Jim Nielsen Published: 2026-10-21T16:51:17.567850Z Content type: opinion Language: en Sources: [Jim Nielsen's Blog]() Topics: [national security](), [intelligence]() Tags: [ai](), [anthropic](), [cancer](), [corporate](), [cost](), [employees](), [lab](), [mission](), [nuclear](), [nuclear-weapons](), [rand-corporation](), [retirement](), [world]() ### AI overview The author compares Daniel Ellsberg's account of RAND during the missile gap years with rhetoric associated with AI labs such as Anthropic. Ellsberg described a close-knit, almost religious sense of mission: RAND researchers saw themselves as insiders with special knowledge, working to save the world from nuclear danger. He and Alain Enthoven skipped RAND's generous retirement plan because, in their late twenties, they doubted they would live to collect it. The author connects this to reports of AI researchers smoking in the belief that AI may cure cancer before smoking harms them, then speculates that ending employer retirement benefits could serve as a signal of faith in near-term universal high income, while also saving companies money. The article excerpt presents this as a provocative analogy and speculation, not a reported policy change. ### Source excerpt I've mentioned before my interest in the history and development of nuclear weapons. Currently, I'm reading The Doomsday Machine: Confessions of a Nuclear War Planner by Daniel Ellsberg (of Pentagon Papers renown). In the first chapter, Ellsberg talks about his time working at the RAND Corporation during the missile gap years when the U.S. thought it was falling behind the Soviet Union in nuclear capabilities. Many of his descriptions sound strikingly similar to the rhetoric coming out of places like Anthropic. For example, note the religious overtones of working at a corporation: I loved RAND, where I ended up spending ten years [...] Much, I imagined, like members of a religious order would, I shared with my colleagues a sense of brotherhood, living and working with others for a transcendent cause. They were insiders: burdened with the responsibility of knowledge others did not possess. precisely in my early missile-gap years at RAND and as a consultant in Washington, there was our sense of mission, the burden of believing we knew more about the dangers ahead, and what might be done about them, than did the generals in the Pentagon or SAC, or Congress or the public, or even the president. It was an enlivening burden. And what was their "enlivening burden"? Saving the world: [My colleagues] shared a feeling--soon transmitted to me--that we were in the most literal sense working to save the world. What really stood out was this part, where Ellsberg describes how he and his colleagues truly believed the end was nigh: From the Air Force intelligence estimates I was newly privy to, and the dark view of the Soviets, which my colleagues shared with the whole national security community, I couldn't believe that the world would long escape nuclear holocaust. Alain Enthoven and I were the youngest members of the department. Neither of us joined the extremely generous retirement plan RAND offered. Neither of us believed, in our late twenties, we had a chance of collecting on i [Next page]()