# cars Published articles for cars. This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles. ## Breaking the Privacy Promise: Gradient Inversion Attacks on Federated Low-Rank Adapted Vision Transformers DevFeed: [Breaking the Privacy Promise: Gradient Inversion Attacks on Federated Low-Rank Adapted Vision Transformers]() Original publisher: [Read original article]() Author: Ritik K. Badiya, Siladittya Manna, Ashmit Sinha, M Yashwanth, Anirban Chakraborty Published: 2026-11-09T05:00:00Z Content type: article Language: en Sources: [arXiv: Cryptography and Security](), [arXiv: Distributed Computing](), [arXiv: Machine Learning]() Topics: [lora](), [Federated Learning](), [peft](), [Computer vision](), [Image processing](), [Cybersecurity]() Tags: [abstract](), [act](), [also](), [artifacts](), [arxiv](), [attacks](), [based](), [batch](), [blocks](), [can](), [cars](), [cifar-10](), [client](), [cs-ai](), [cs-cr](), [cs-cv](), [cs-dc](), [cs-lg](), [federated](), [gradient](), [privacy](), [rank](), [transformers](), [vision](), [vision-transformers]() ### AI overview The study finds that LoRA fine-tuning alone does not prevent private image leakage from federated Vision Transformers when a server can access individual client updates. It attributes reconstruction ambiguity to weak low-rank gradient signals and missing spatial cues from frozen positional embeddings. The proposed framework adds server-prescribed trainable positional embeddings during client fine-tuning; their gradients provide layer-wise spatial anchors, while a generative prior reduces duplicate-data and patch-permutation artifacts. When added to GIAS, this approach improves reconstruction metrics in FedSGD: on CIFAR-10, PSNR rises 10.14 dB, SSIM 0.45, and LPIPS improves by 0.098; PSNR gains are 7.83 dB on CelebA and 11.74 dB on CARS. The abstract reports that gains persist at higher batch sizes and with FedAvg. ### Source excerpt arXiv:2611.09652v1 Announce Type: cross Abstract: Parameter-efficient fine-tuning (PEFT) has been increasingly adopted to fine-tune pretrained models with low computational cost and is therefore also preferred in the federated learning paradigm to reduce computational and communication overhead. In this work, we investigate the sparsely explored problem of the risk of private data leakage in LoRA-adapted Vision Transformers (ViTs), where low-rank updates weaken the gradient signal available to a server with access to individual client updates. We show that quality of reconstruction is limited by two coupled ambiguities: reduced information in LoRA gradients and missing spatial cues from frozen positional embeddings. To address this, we propose a spatially anchored generative inversion framework that incorporates server-prescribed trainable positional embeddings into ViT blocks during fine-tuning at clients. The resulting positional gradients act as layer-wise spatial anchors, providing additional positional cues, while the generative prior keeps reconstructions on a realistic image manifold, reducing duplicate-data and patch-permutation artifacts, {which produce similar gradients as those produced when the target client images are used}. Adding our proposed spatial anchoring to GIAS improves PSNR by 10.14 dB, SSIM by 0.45, and LPIPS by 0.098 on CIFAR-10, as well as PSNR gains of 7.83 and 11.74 dB on high-resolution CelebA and CARS datasets, respectively, in the FedSGD setting. Such gains in reconstructed image quality metrics are also maintained at higher batch sizes and in FedAvg settings. These results indicate that LoRA-based PEFT does not, by itself, prevent data leakage in FL, and carefully incorporated positional embeddings can help in exposing private information. ## Digital billboards across the globe hacked to show porn, attackers say "anyone with two brain cells" could do it DevFeed: [Digital billboards across the globe hacked to show porn, attackers say "anyone with two brain cells" could do it]() Original publisher: [Read original article]() Author: Rob Thubron Published: 2026-11-02T14:09:00Z Content type: news Language: en Sources: [TechSpot]() Topics: [Digital signage](), [Cybercrime]() Tags: [2019](), [2025](), [2026](), [accessible](), [also](), [asics](), [australia](), [automated](), [basic](), [begun](), [broadcast](), [cars](), [digital](), [drivers](), [egypt](), [hacked](), [hacking](), [porn](), [screen](), [security](), [show](), [tech-culture](), [text]() ### AI overview Digital billboards in several countries were reported to have shown pornography after attackers allegedly exploited a vulnerability in software used to send content to LED displays. The group Leakers$ claimed an automated scan found thousands of exposed systems in about two hours, accessible through web pages and interfaces lacking basic protections, and said its aim was to expose weak security. Egypt's Interior Ministry attributed the incident to a cyberattack on an international company contracted by the billboard operator; Al Arabia United shut down an affected screen and began investigating with authorities. Malaysia's Kluang Municipal Council also switched off compromised screens and said it was working with its vendor to investigate and improve security. Brazil was confirmed in the report; incidents in Iraq and China were described as claims. Legal proceedings in Egypt had begun. ### Source excerpt The most visible hack took place on a digital billboard stretching across one of Egypt's biggest highways: the 26th of July Axis in Giza. Drivers stopped their cars to film as the screen broadcast the adults-only material, which was overlaid with the text "Leaked by: Ch, 19ylegal never and nivex"... Read Entire Article ## Bloatware in Aftermarket Auto Parts DevFeed: [Bloatware in Aftermarket Auto Parts]() Original publisher: [Read original article]() Author: Bruce Schneier Published: 2026-10-30T15:20:02Z Content type: news Language: en Sources: [Schneier on Security]() Topics: [Android Security](), [WebView](), [Web]() Tags: [adobe](), [ai](), [amazon](), [android](), [baidu](), [call](), [cars](), [claude](), [dab](), [internet-of-things](), [iot](), [pdfs](), [power](), [radio](), [reverse-engineering](), [smart-tv](), [software](), [spyware](), [system](), [uncategorized](), [voice-assistant](), [webview]() ### AI overview A Civic owner used Claude to examine a cheap Amazon aftermarket head unit and found several preinstalled apps: a voice assistant using Android WebView, Baidu location services, a call recorder, a DAB radio client, and Adobe Acrobat. The voice assistant reportedly consumed about a third of the device's processing power even while idle. The excerpt presents this as an example of software and security concerns in aftermarket automotive electronics, and notes that AI helped an individual reverse-engineer the device. ### Source excerpt This story is interesting: One Civic owner purchased a cheap aftermarket head unit off Amazon, and decided to have Claude work its magic on it, like it would a smart TV. What was reportedly discovered isn't surprising, but is worth knowing about. Among the pre-installed bloatware--or spyware, if you're being less charitable--was a voice assistant equipped with WebView, an Android system component that lets apps display content off the web. This apparently always used about a third of processing power, even while idle, which isn't great. There were also Baidu location services, a call recorder (!), a DAB radio client that is pretty much useless in the U.S., and Adobe Acrobat, presumably so your passenger can entertain themselves with PDFs on the go... ## Can AI data centers be mass-produced like cars? This startup aims to find out DevFeed: [Can AI data centers be mass-produced like cars? This startup aims to find out]() Original publisher: [Read original article]() Author: Dominic-Madori Davis Published: 2026-10-30T13:00:00Z Content type: news Language: en Sources: [TechCrunch | AI News & Artificial Intelligence]() Topics: [data centers](), [clean energy data centers]() Tags: [ai](), [ai-data-centers](), [ai-deployment](), [ai-factories](), [ai-infrastructure](), [air](), [amazon](), [assemble](), [assembly](), [automotive](), [automotive-industry](), [build](), [building](), [cars](), [data-centers](), [factory](), [runware](), [scale](), [startup](), [startups](), [test]() ### AI overview Runware plans to mass-produce prefabricated AI data center pods at a Romania factory, targeting 30 megawatts of capacity per day by 2028. The company says a pod can be installed and set up within 48 hours; as production ramps, it expects pod construction to fall from three weeks to one day. Customers may provide GPUs or rent compute, while power and internet are charged separately. Runware says pods use closed-loop liquid cooling with air-cooled radiators and can be configured for customer needs and new GPU generations. These are plans whose execution remains unproven: scaling will require staff, inventory, logistics and funding, and some components come from Germany while networking depends on Nvidia. The excerpt also reports Runware's claimed customer interest and planned factory expansion, alongside its stated efforts to use renewable power. ### Source excerpt Runware is building a factory to turn out prefabricated AI data centers on an assembly line. The real test will be whether it can deliver at that scale. ## AI in Automotive: Today's Breakthroughs and Tomorrow's Roadmap DevFeed: [AI in Automotive: Today's Breakthroughs and Tomorrow's Roadmap]() Original publisher: [Read original article]() Author: CoreWeave Published: 2026-10-22T16:27:39Z Content type: article Language: en Sources: [Coreweave Blog Posts]() Topics: [Artificial Intelligence](), [AI Infrastructure](), [AI Integration Strategies](), [Advanced AI workflows]() Tags: [2](), [2025](), [3](), [about](), [accelerate](), [adaptive](), [adas](), [advanced](), [advanced-driver-assistance](), [advice](), [ai](), [ai-applications](), [ai-cloud](), [ai-development](), [ai-infrastructure](), [automotive](), [cars](), [content-type](), [coreweave](), [evs](), [roadmap]() ### AI overview The interview identifies ADAS, predictive maintenance, and EV energy management as near-term automotive AI uses, with AI also expected to shape in-car experiences, manufacturing, and supply chains. It says organizations struggle to scale training, control experimentation costs, and move models into deployment because training, low-latency inference, and simulation need different infrastructure. The interviewee argues that AI-specialized cloud infrastructure can offer more elastic, workload-focused support than generalized hyperscalers, while DIY private clouds offer control at the cost of agility and scale. For investment, the interview recommends combining in-car differentiation with manufacturing efficiencies, which may deliver faster measurable returns, and integrating AI across vehicle, factory, and lifecycle operations. These are CoreWeave's claims and outlook, not independent comparisons or measured results. ### Source excerpt AI is transforming cars, EVs, and factories. See CoreWeave's take on today's breakthroughs and tomorrow's automotive roadmap. ## Tesla claims privately owned Model Ys can join its Robotaxi fleet in Q2 2027 - a deadline it will likely miss DevFeed: [Tesla claims privately owned Model Ys can join its Robotaxi fleet in Q2 2027 - a deadline it will likely miss]() Original publisher: [Read original article]() Author: Daniel Sims Published: 2026-10-22T11:33:00Z Content type: news Language: en Sources: [TechSpot]() Topics: [autonomous vehicles]() Tags: [2020](), [2025](), [2026](), [austin](), [autonomous](), [autonomous-driving](), [autonomous-vehicles](), [autonomus](), [car-tech](), [cars](), [ceo](), [crashes](), [drivers](), [earnings](), [electric-cars](), [electronics](), [elon-musk](), [fleet](), [fleet-management](), [model](), [robotaxi](), [self-driving](), [tesla](), [third-quarter](), [vehicles]() ### AI overview Tesla says some Model Y owners may be able to add their cars to its Robotaxi fleet around Q2 2027, potentially earning ride-hailing revenue. The cars would need the autonomous software planned for V15 and integration with Tesla's geofenced operating areas and fleet-management system. The excerpt argues the deadline may slip: Tesla's robotaxi launch came years after Elon Musk's earlier 2020 target, and its service remains limited, with some cars still carrying human supervisors. It also reports that Tesla robotaxis have recorded more crashes per mile than typical human drivers. The article contrasts Tesla's vision-only sensing approach with Waymo and Zoox, which also use radar and lidar. Despite the slow rollout, more than three-quarters of new North American Tesla buyers chose Full Self-Driving (Supervised) last quarter. ### Source excerpt During Tesla's third-quarter earnings call on October 21, VP for Autonomous Driving Ashok Elluswamy announced the company's aim to let some Model Y owners add their vehicles to Robotaxi's autonomous fleet around the second quarter of 2027. The feature, which would enable users to earn money from ride-hailing, has long... Read Entire Article ## When Thoughts Become Memory: Unifying In-Context Learning and Fine-Tuning DevFeed: [When Thoughts Become Memory: Unifying In-Context Learning and Fine-Tuning]() Original publisher: [Read original article]() Author: Leon Chlon, Julie Huang Published: 2026-10-22T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Machine learning](), [Fine-tuning](), [Enhancing AI model performance](), [context window]() Tags: [abstract](), [airfoil](), [architecture](), [arxiv](), [beats](), [better](), [bits](), [cache](), [camera](), [cars](), [certificate](), [code](), [comparisons](), [context](), [cs-lg](), [data](), [dataset](), [decoding](), [fine-tuning](), [learning](), [memory](), [stat-ml]() ### AI overview The paper proposes a code-length rule for deciding whether data are more useful in a model's context or weights: value is the expected code length saved on held-out queries, related to information stored about the query, prior repair, and decoding loss. It says upper bounds can certify a choice from one measurement, while an anytime-valid held-out test controls error whenever it stops. Reported comparisons include TabPFN over XGBoost and a trained surrogate over an in-context learner for airfoil pressure; for GPT-2, the test selected the better of cache and fine-tuning in 61,948 of 64,000 query streams across 32 comparisons, with at most 3.3% errors on null streams. Context ablations are presented as held-out validation: in an occupancy network, one patch seeing a shape feature reduced local code length a hundredfold, whereas eight patches missing it nearly doubled it. The abstract also reports a stereo-network advantage for one camera seeing the surface over four, and AUROC 0.91 when the valuation recovered a published evidence adjudicator. ### Source excerpt arXiv:2610.29419v1 Announce Type: cross Abstract: We are the first to give a certified rule for whether unseen data belong in a pretrained model's context or its weights, built on an exact law for their information value. That value, the code length saved on held-out queries, equals in expectation the bits stored about the query, plus a prior repair, minus a decoding loss, for any model, architecture, dataset and task. The law's ceilings certify the choice from one measurement, and an anytime-valid test makes it from held-out queries, with error controlled whenever it stops. Every certificate the ceilings issue on a trained transformer, TabPFN, an occupancy network and a stereo network is correct. The test picks TabPFN over XGBoost and a trained surrogate over an in-context learner on airfoil pressure, calls the better of GPT-2's cache and fine-tune in 61,948 of 64,000 query streams over 32 comparisons, and errs in at most 3.3% of null streams. The law also diagnoses when learning in context helps and when it does not. Deleting part of a context is held-out validation of that part in any model, and adding parts one at a time traces the learning curve in context: one patch that sees a shape feature cuts an occupancy network's code length there a hundredfold, while eight that miss it nearly double it, and one camera that sees the surface beats four in a stereo network. Ranked by this label-free valuation, a kriging learner's three best support cars beat four drawn at random, and on a yes-or-no claim the valuation recovers a published evidence adjudicator, with AUROC 0.91. ## To Keep or Not to Keep: Learning KV Cache Retention in Disaggregated LLM Serving Systems DevFeed: [To Keep or Not to Keep: Learning KV Cache Retention in Disaggregated LLM Serving Systems]() Original publisher: [Read original article]() Author: Dong Liu, Yanxuan Yu, Eric Jiang, Shu Wang, Ying Nian Wu Published: 2026-10-22T04:00:00Z Content type: article Language: en Sources: [arXiv: Distributed Computing](), [arXiv: Machine Learning]() Topics: [LLM optimization](), [Large Language Model](), [Cache](), [Machine learning]() Tags: [adaptive](), [admission](), [bandwidth](), [cache](), [cars](), [controller](), [costs](), [cs-dc](), [cs-lg](), [kv-cache](), [learning](), [llm](), [llm-serving](), [lru](), [memory](), [memory-pressure](), [model-weights](), [multimodal](), [network](), [retention](), [reuse](), [storage]() ### AI overview KVLearn targets disaggregated LLM serving, where a cache miss costs both recomputation and network transfer, while admitting a block also consumes transfer bandwidth and storage. Its Prefix Reuse Predictor estimates reuse from prefix structure and timing without changing model weights; a Cost-Aware Retention Score weighs recomputation, transfer, and storage costs for keep/admit decisions; and an Adaptive Threshold Controller adjusts admission thresholds using observed hit rates and memory pressure. Evaluated on text and multimodal workloads, including image- and video-derived KV blocks, it reduced time-to-first-token by up to 56% versus No-Cache, 38% versus LRU-Pool, and 33% versus Mooncake-style baselines. It cut inter-node KV transfer volume by up to 53% versus LRU-Pool; on MM-Session, throughput stayed within about 5% of oracle. These are reported maxima and workload-specific results. ### Source excerpt arXiv:2610.28637v1 Announce Type: new Abstract: Disaggregated \textsc{LLM} serving separates prefill and decode across node pools, placing a network fabric between \textsc{KV}-cache computation and consumption. This breaks a key assumption of classical cache policies: a miss no longer incurs only local recomputation, but also network transfer to the decode node. Likewise, admitting a block to the global cache requires an additional transfer, so poor retention decisions waste both memory and bandwidth. We present \textbf{KVLearn}, a learning-based framework for cost-aware \textsc{KV}-cache retention in disaggregated serving. KVLearn includes: (i) a lightweight Prefix Reuse Predictor (\textsc{PRP}) that estimates reuse probability from structural and temporal prefix features without modifying model weights; (ii) a Cost-Aware Retention Score (\textsc{CARS}) that converts reuse probability into keep/admit decisions using recomputation, transfer, and storage costs; and (iii) an Adaptive Threshold Controller (\textsc{ATC}) that adjusts admission thresholds online from observed hit rates and memory pressure. We integrate KVLearn into a globally disaggregated serving topology and evaluate it on text and multimodal workloads, including large image/video-derived \textsc{KV} blocks with heterogeneous reuse patterns. KVLearn reduces end-to-end time-to-first-token by up to 56% versus No-Cache, 38% versus LRU-Pool, and 33% versus Mooncake-style baselines, while cutting inter-node \textsc{KV} transfer volume by up to 53% versus LRU-Pool. On MM-Session, throughput remains within ${\sim}5%$ of oracle. Code is available at https://github.com/FastLM/KVLearn. ## Be a loser DevFeed: [Be a loser]() Original publisher: [Read original article]() Author: Sergio Visinoni Published: 2026-10-21T05:00:07Z Content type: opinion Language: en Sources: [Sudo Make Me a CTO]() Topics: [intrinsic motivation](), [Learning](), [Ruby on Rails]() Tags: [2026](), [ai](), [ai-assisted](), [ai-assisted-coding](), [brain](), [car](), [cars](), [circuit](), [claude](), [claude-code](), [code](), [coding](), [craft](), [dhh](), [keynote](), [leadership](), [rails](), [rails-world-2026](), [software](), [software-engineers](), [warning]() ### AI overview The author argues that software engineers can keep writing code for craftsmanship and enjoyment even if AI can produce it faster, and criticizes DHH's "Don't be a loser" message as pressure without compelling justification. The article challenges the speech's narrow economic view of coding through comparisons with painting and car racing, and says its focus on speculative AI doom overlooks present harms the author identifies, including environmental impacts, worker exploitation, skill atrophy, and workplace distress. It also criticizes DHH's stated willingness to benefit from Rust code generated by AI while avoiding direct engagement with Rust, framing this as part of a broader pattern of separating benefits from consequences. The author urges readers to resist pressure to adopt AI, pursue their craft, and continue learning. ### Source excerpt DHH's keynote at Rails World 2026 ended with a warning for all software engineers "Don't be a loser". Here's a warning about listening to people like him without turning on your brain. ## HullProbe: Measuring Interventional Part Reliance in Maritime Vessel Classifiers DevFeed: [HullProbe: Measuring Interventional Part Reliance in Maritime Vessel Classifiers]() Original publisher: [Read original article]() Author: Faisal Mohammad, Mohamed Younis Published: 2026-10-21T04:00:00Z Content type: article Language: en Sources: [arXiv: Machine Learning]() Topics: [Artificial Intelligence](), [Image](), [test]() Tags: [annotations](), [architectures](), [audit](), [cars](), [classifiers](), [cs-cv](), [cs-lg](), [dataset](), [fine](), [generic](), [grid](), [influence](), [interaction](), [maritime](), [marvel](), [measuring](), [monitoring](), [parts](), [proxy](), [type]() ### AI overview HullProbe measures which vessel parts drive a classifier by exhaustively swapping regions in a 2x3 hull-and-superstructure grid, using same-class swaps to control for compositing artifacts and a Walsh transform to estimate each part's influence and interactions. In planted-shortcut tests, its measure tracked worst-group accuracy (r = −0.94 across 45 configurations); human annotations changed the estimated hull share by less than the audit's resolution. On MARVEL's 26 vessel types, the classifier attributed about 46% of corrected sensitivity to the generic hull, but followed the superstructure when the parts conflicted. Available group-label-free debiasers did not detectably lower hull reliance. The abstract reports similar findings across three architectures and another maritime dataset, and an extension to Stanford Cars. ### Source excerpt arXiv:2610.25137v1 Announce Type: cross Abstract: Recognizing a vessel's type from an image supports maritime surveillance and sanctions monitoring, but a classifier can reach high accuracy by leaning on the generic hull instead of the more distinctive superstructure, and is then fooled when a vessel is re-hulled or occluded. De-biasing methods claim to remove such reliance, yet the field certifies them with worst-group accuracy, a proxy that needs group labels and only infers reliance from accuracy. We measure the reliance directly. HullProbe swaps a registered anatomical part of a vessel for another instance's and reads how much that part drives the decision, using a same-class swap to control for compositing artifacts. Because a surface vessel is a hull and a superstructure arranged along one axis, its parts form a 2x3 grid whose 2^6 = 64 swaps we evaluate exhaustively, and a single Walsh transform returns each part's interventional influence and its interaction spectrum. On planted shortcuts, where the true reliance is known, the measure tracks worst-group accuracy (r = -0.94 over 45 configurations), and replacing its learned segmenter with human annotations changes the hull-share by less than the audit's resolution. On the 26 vessel types of MARVEL, the classifier places about 46% of its corrected sensitivity on the generic hull yet follows the superstructure when the two conflict, and the group-label-free de-biasers available there do not detectably reduce the hull's share. The findings hold across three architectures and a second maritime dataset, and the method extends to fine-grained vehicle recognition on Stanford Cars. ## End-to-End AI vs. Traditional Self-Driving Systems DevFeed: [End-to-End AI vs. Traditional Self-Driving Systems]() Original publisher: [Read original article]() Author: vishwa vimukthi gammuduwaththage Published: 2026-10-19T14:45:51Z Content type: article Language: en Sources: [HackerNoon]() Topics: [autonomous vehicles](), [AI agents in autonomous vehicles](), [Artificial Intelligence](), [World models](), [Physical AI](), [LLM security]() Tags: [ai](), [autonomous-driving](), [autonomous-vehicle-regulation](), [autonomous-vehicle-safety](), [autonomy](), [av](), [cars](), [debugging](), [driving](), [end-to-end-autonomous-driving](), [neural-network-explainability](), [nvidia-alpamayo](), [safety](), [self-driving](), [systems](), [tesla](), [tesla-cybercab](), [vision-language-action-models](), [waymo](), [wayve-ai-driver]() ### AI overview The article compares modular and end-to-end systems for autonomous driving, focusing on debugging, validation, safety oversight, regulation, and sensing and mapping choices. It argues that deployed systems combine learned drivers with retained structure, independent safety layers, and evidence to support safety claims. ### Source excerpt Tesla, Wayve, and Waymo are advancing AI-driven cars. Explore the challenges of debugging neural drivers, verifying safety, and regulating autonomy. ## AI agents are coming to Rivian's cars DevFeed: [AI agents are coming to Rivian's cars]() Original publisher: [Read original article]() Author: Sean O'Kane Published: 2026-10-15T15:02:36Z Content type: news Language: en Sources: [TechCrunch | AI News & Artificial Intelligence]() Topics: [AI agents in autonomous vehicles](), [Physical AI](), [Local AI](), [human in the loop agent](), [agent orchestration](), [Plover]() Tags: [agentic-ai](), [ai](), [ai-agents](), [ai-autonomy](), [anthropic](), [autonomy](), [cars](), [electric-vehicles](), [evs](), [frontier-ai](), [implement](), [implementation](), [llms](), [rivian](), [rj-scaringe](), [techcrunch-disrupt](), [transportation]() ### AI overview Rivian plans to let drivers use AI agents from companies such as Anthropic in its vehicles, building on an in-house assistant that routes requests among local and cloud models and can access external apps. The company has not announced a timeline. It says integrating outside agents will require coordination and safeguards, with vehicle controls such as steering and braking kept out of agents' hands. ### Source excerpt We'll likely learn more about how this will work at the company's upcoming "AI & Autonomy Day" event in December. ## Reverse-Engineering Flock Cameras DevFeed: [Reverse-Engineering Flock Cameras]() Original publisher: [Read original article]() Author: Bruce Schneier Published: 2026-09-21T14:37:45Z Content type: article Language: en Sources: [Schneier on Security]() Topics: [Reverse Engineering](), [Computer vision](), [Security](), [Encryption](), [Software]() Tags: [ai](), [alpr](), [cameras](), [cars](), [computer-vision](), [data](), [encryption](), [flock](), [reverse-engineering](), [security](), [security-engineering](), [software](), [uncategorized]() ### AI overview A reverse-engineering analysis of a captured Flock automatic license plate reader camera found software that detects people, vehicles, license plates, bicycles, bumper stickers, and other graphics. Recovered logs indicate the device generated more than a million images, while an unencrypted partition contained a key for an encrypted partition. ### Source excerpt Hackers captured a Flock camera and got a look (alternate link) at the software: While much of the automatic license plate reader's (ALPR) most sensitive storage remained encrypted and inaccessible, the joint analysis of the recovered data shows that software running on the device explicitly detects people as well as vehicles, license plates, and bicycles. The camera can produce dozens of images of a single passing vehicle and, according to several weeks of recovered logs, generated more than a million images. Its computer-vision software also sometimes isolated bumper stickers and other graphics, including, in one case, an American flag patch on a motorcyclist's saddlebag... ## The French and Germans will ban Tesla FSD to protect their car industry DevFeed: [The French and Germans will ban Tesla FSD to protect their car industry]() Original publisher: [Read original article]() Author: Pieter Levels Published: 2026-07-22T00:00:00Z Content type: opinion Language: en Sources: [@levelsio (Pieter Levels)]() Topics: [autonomous vehicles]() Tags: [ban](), [car](), [cars](), [customers](), [driving](), [eu-acc](), [europe](), [evs](), [france](), [fsd](), [full-self-driving](), [gdpr](), [industry](), [policy](), [politics](), [privacy](), [required](), [self-driving](), [tech](), [tesla]() ### AI overview The author argues that France and Germany may ban Tesla FSD to protect their auto industries from competition, claiming local cars lack comparable self-driving features and customers could flock to Tesla. The excerpt offers no evidence that a ban is planned or explains the regulatory situation; it ends with a joke about a GDPR privacy notice printed on a car. ### Source excerpt The French and Germans will ban Tesla FSD to protect their car industry that's being decimated by Chinese EVs and Tesla Allowing Tesla full self driving would ... ## Inside the Geofence: Faster Pickups Start Here DevFeed: [Inside the Geofence: Faster Pickups Start Here]() Original publisher: [Read original article]() Author: Global Technology Published: 2026-05-12T13:14:04Z Content type: article Language: en Sources: [McDonald]() Topics: [geofencing](), [Software Testing](), [Mobile Testing](), [Test automation](), [digital experience monitoring]() Tags: [actions](), [applications](), [approach](), [automation](), [based](), [behavior](), [can](), [capture](), [cars](), [cellular](), [challenge](), [critical](), [customer](), [customer-experience](), [customers](), [data](), [digital](), [don-t](), [faster](), [geofencing](), [here](), [inside](), [mobile](), [technology](), [testing]() ### AI overview McDonald's uses geofencing-enabled field tests to check how mobile order features respond as customers approach and move through real restaurant settings. GPS, Wi-Fi, and cellular data define virtual zones; layered perimeters can trigger a notification, start kitchen preparation, and confirm pickup arrival. Testing drive-thru, curbside, counter, and dine-in journeys across varied locations helps expose timing, accuracy, connectivity, device, and usability problems that controlled lab tests may miss. For Ready on Arrival, the app alerts crews when a customer is nearby so they can prepare orders just in time; the excerpt says the feature launched in US markets in April 2023 and expanded to many large markets. The approach logs event telemetry and supports device and app-version comparisons. Dense urban areas can cause missed or false triggers; proposed mitigations include Wi-Fi triangulation and adjustable geofence sizes. The excerpt also flags privacy compliance, device variability, and store training and workflow readiness as concerns. ### Source excerpt Geofencing field testing turns noisy GPS signals into reliable arrivals for drive-thru, curbside, and beyond. By Chandrani Gupta Chowdhury, Sr. Manager, Quality Assurance Engineering Quick Bytes: Lab testing alone can't fully capture the real-world conditions that shape mobile experiences inside McDonald's restaurants As customers drive toward a restaurant, pull into a parking spot, or move through a drive-thru lane, geofencing-enabled field testing shows how digital features respond in real time This approach improved accuracy, timing, and reliability -- strengthening customer experiences like Ready on Arrival (ROA), curbside, and loyalty Mobile experiences don't happen in perfect conditions -- they happen in cars, parking lots, drive-thru lanes, and busy restaurants where signals fluctuate and customer behavior is unpredictable. For McDonald's, ensuring digital features work reliably in these real-world environments is critical to delivering fast, seamless experiences at scale. To meet that challenge, teams began integrating geofencing into field testing, allowing them to observe how mobile features perform as customers approach, arrive, and move through restaurant spaces. Inside the digital boundary of the restaurant Geofencing is a location-based technology that leverages GPS, Wi-Fi, and cellular data to create virtual boundaries around physical locations, enabling applications to trigger predefined actions when users enter or exit those zones. The benefits of incorporating geofencing into field testing are significant. From an operational standpoint, it helps ensure restaurants are prepared to receive and fulfill digital orders efficiently, reducing customer wait times and improving order accuracy. From a customer experience perspective, geofencing enables timely, relevant, and seamless interactions, reinforcing trust in the digital journey. Within the McDonald's ecosystem, geofencing can be applied across restaurant footprints -- including parking areas and drive ## Now in Android #117 -- Google I/O 2025 Part I DevFeed: [Now in Android #117 -- Google I/O 2025 Part I]() Original publisher: [Read original article]() Author: Daniel Galpin Published: 2025-06-05T17:28:56Z Content type: news Language: en Sources: [Android Developers]() Topics: [Material Design](), [Android](), [Design system](), [software-development](), [User Experience](), [Large Screen](), [3D]() Tags: [2025](), [adaptive](), [ai](), [android](), [android-16](), [android-xr](), [animations](), [cars](), [cloud](), [colors](), [design](), [design-patterns](), [development](), [displays](), [featured](), [gemini](), [google-i-o](), [google-io-2025](), [guidelines](), [i-o](), [laptops](), [material](), [now-in-android]() ### AI overview Google I/O 2025 introduced Material 3 Expressive, with a new motion physics system, typography for variable and static fonts, morphing shapes, more colors, and expanded configuration options across 15 new or updated components. The Q3 Android 16 beta includes much of this visual polish and is available for supported Pixel devices. Google also released the Wear OS 6 developer preview, which lets developers test apps in its emulator. The excerpt additionally points to design guidance for accessibility and multiple Android form factors, but does not provide the rest of the post's coverage. ### Source excerpt Now in Android #117 -- Google I/O 2025 Part IDesign, Watches, Cars, Tablets, Adaptive Apps, Android XR, Gemini, and Android 16 Welcome to part one of a special two-part Google I/O 2025 edition of Now in Android. This first post will cover a bunch of changes related to the latest evolution of Material Design, watches, cars, tablets, laptops, and connected displays, the latest in adaptive app development, XR development, how to take advantage of on-device and cloud based AI, and Android 16. https://medium.com/media/f4779f9096112bf2cfff21aec7eeb199/hrefhttps://medium.com/media/d23b17afda66e01541bb4c84ecea4974/href Most of the content of this post is available in the form of a video or podcast, so feel free to watch or listen rather than read on. (Or do all three to help you remember! There won't be a quiz.) The Android Show: I/O Edition -- what Android devs need to know! 🤖 We began the I/O season with a special edition of The Android Show, where we introduced the latest evolution of Material Design, Material 3 Expressive. https://medium.com/media/815112dc9a5b3a849cced5343837f4e6/href The Android Show: I/O Edition - what Android devs need to know! https://medium.com/media/906cc031e2cbb814b2a4964db156a2cc/href Material 3 Expressive adds a new motion physics system, new type styles for variable and static fonts, an expanded shape library with morphing animations, and an expanded range of colors. Fifteen new or updated components now feature more configuration capabilities, shape options, emphasized text, and other expressive updates. The Material team has a post where you can read all about it, including design tactics. M3 Expressive: Engaging UX Design At I/O Build next-level UX with Material 3 Expressive covered how to use the new expressive design patterns; breaking down the research, explaining new guidelines, and including new design files + code. https://medium.com/media/4ce31edc8b396f71acc6192d1ac6160f/href The first beta of the Q3 Android 16 update contains much of ## Self driving cars react 150x faster than a human driver DevFeed: [Self driving cars react 150x faster than a human driver]() Original publisher: [Read original article]() Author: Pieter Levels Published: 2024-12-18T00:00:00Z Content type: comparison Language: en Sources: [@levelsio (Pieter Levels)]() Topics: [autonomous vehicles]() Tags: [ai](), [automation](), [cars](), [driver](), [driving](), [execution](), [faster](), [future](), [reaction](), [self-driving](), [self-driving-car](), [self-driving-cars](), [tech](), [time]() ### AI overview The text says self-driving cars react and execute faster than human drivers: it lists a human reaction time of 0.3 seconds and execution time of 1-2 seconds, compared with 0.001 seconds and 0.1 seconds for a self-driving car. These figures support the headline's broad speed comparison, though the excerpt does not explain how the times were measured. ### Source excerpt Self driving cars can react 150x faster than a human driver Human driver: reaction time 0.3 seconds execution time 1-2 seconds Self driving car: reaction time ... ## Volvo drives into the future of online car shopping with the supergraph DevFeed: [Volvo drives into the future of online car shopping with the supergraph]() Original publisher: [Read original article]() Author: Melissa Annecchini Published: 2022-12-01T16:48:44Z Content type: article Language: en Sources: [Apollo Blog]() Topics: [Apollo Connectors](), [UI/UX Design](), [Web]() Tags: [2025](), [aggregates](), [api](), [apis](), [apollo](), [car](), [cars](), [community](), [complex](), [developers](), [experience](), [graphics](), [it](), [journey](), [online](), [pricing](), [sales](), [shopping](), [system](), [user-experience](), [website]() ### AI overview Volvo Cars adopted a supergraph with Apollo GraphOS after its GraphQL monolith became difficult to scale: changes took a long time, website experiences broke, and a central team had to troubleshoot. The supergraph lets teams own and manage their data through subgraphs, while GraphOS shows which APIs are in use so teams can assess potential breaking changes. The excerpt reports that unplanned breaking changes were eliminated and that frontend developers found needed data 99.5% faster after the supergraph unified data from 40 REST endpoints. The excerpt does not provide further detail on the online sales goal or the full implementation. ### Source excerpt Volvo Cars is on a journey to have at least 50% of their sales online by 2025. To achieve this goal, the organization needed to create a consistent and intuitive user experience across its website. However, Volvo's system landscape was so complex that it was taking developers over a week to figure out how to combine APIs into new experiences. To complicate matters further, there were site delays that caused abandonment and shoppers weren't seeing accurate pricing information or graphics. ## Interpreting voice results for Android media apps in cars DevFeed: [Interpreting voice results for Android media apps in cars]() Original publisher: [Read original article]() Author: Danny Preussler Published: 2022-01-10T21:33:42Z Content type: tutorial Language: en Sources: [Danny Preussler]() Topics: [Android Auto](), [Android](), [media3](), [Android Testing]() Tags: [actions](), [android](), [android-app-development](), [android-auto](), [androiddev](), [cars](), [google-assistant](), [integration](), [media](), [voice]() ### AI overview For Android Auto media searches, the excerpt reports that Assistant calls `MediaSessionCompat.Callback.onPlayFromSearch()` with the remaining query, but the documented `MediaStore.EXTRA_MEDIA_FOCUS` check does not distinguish artist, album, or track: it says the received focus is always `vnd.android.cursor.item/*`. The suggested workaround is to read the non-null `EXTRA_MEDIA_ARTIST`, `EXTRA_MEDIA_ALBUM`, and `EXTRA_MEDIA_TITLE` values from the callback bundle to identify the requested entities. The author notes the cause is unclear and recommends testing the implementation, since documented behavior may not match reality. ### Source excerpt When working on the Android Auto integration for SoundCloud, I stumbled upon an interesting issue. When it comes to implementing voice actions, things you read in the documentation might differ a bit from reality. Voice commands are very critical while driving a car. You want the driver to keep looking at the street, keep his hands on the steering wheel, instead of interacting with any screen. This is why when building for Android Auto, Automotive OS, and Assistant Driving Mode you have to support basic voice searches. For a music app like ours, the idea is simple. A user can say something like: "Play Moderat on SoundCloud" Under the hood The Assistant will break down our sentence for interpretation: " " The verb (Play) and the app (SoundCloud) in the above command were meant for the system. It will then wake up the app mentioned (if it declared auto support via manifest) and, as we asked for verb "Play", the method MediaSessionCompat.Callback.onPlayFromSearch() will be called with the remaining query ("Moderat") as the argument. This query of "what to play" can still mean different things though. How do you know if Moderat is a band or a song? Don't worry, the Assistant will help us also here. It already got some idea what our query could mean, if we asked for an artist, an album, or a specific track. But be aware, it behaves slightly differently than officially documented. According to documentation we are supposed to write code like this: val mediaFocus = extras?.getString(MediaStore.EXTRA_MEDIA_FOCUS) if (mediaFocus == MediaStore.Audio.Artists.ENTRY_CONTENT_TYPE) { isArtistFocus = true artist = extras.getString(MediaStore.EXTRA_MEDIA_ARTIST) We are suppose to check for a bundle entry with the key MediaStore.EXTRA_MEDIA_FOCUS . This will give us a hint on how best to interpret the query. The value of the constant MediaStore.EXTRA_MEDIA_ARTIST checked above is vnd.android.cursor.item/artist and there are similar constants related to ## Release Radar - July 2018 DevFeed: [Release Radar - July 2018]() Original publisher: [Read original article]() Author: Lee Reilly Published: 2018-08-17T07:00:00Z Content type: news Language: en Sources: [GitHub Blog]() Topics: [GigaOm Radar](), [Open Source](), [User interface design](), [Game Off](), [Node.js](), [Testcontainers]() Tags: [ai](), [android](), [apollo](), [application](), [bsd](), [cars](), [code](), [database](), [news-insights](), [open-source](), [radar](), [release-radar](), [the-library]() ### AI overview A roundup of software and open source releases from July 2018 highlights tools for learning programming, linting, content management, customer support, application inspection, autonomous driving, command-line interfaces, and game development. The featured updates include raylib 2.0, Mu 1.0, ALE 2.0, Pico 2.0, Helpy 2.0, Reactotron 2.0, Apollo 3.0, Ora 3.0, and Xenko becoming free and open source. ### Source excerpt Open source releases that caught our attention last month. ## Great Things Take Time DevFeed: [Great Things Take Time]() Original publisher: [Read original article]() Published: 2013-03-04T15:41:57Z Content type: opinion Language: en Sources: [Michael Dubakov]() Topics: [software-development](), [CI/CD](), [The Three Main Cloud Service Models](), [upgrade]() Tags: [apple](), [application](), [applications](), [cars](), [change](), [company](), [complexity](), [created](), [deep](), [developers](), [development](), [fail](), [fast](), [faster](), [game](), [game-development](), [games](), [good](), [history](), [time]() ### AI overview The essay argues that meaningful product success often takes time: software can be updated and iterated quickly, but sustained focus and repeated attempts help produce better solutions. It contrasts "hedgehog" companies, which concentrate on one area, with "fox" companies that pursue varied markets and projects. Rovio's many games before Angry Birds, Apple's careful product launches, and Salesforce's enterprise SaaS focus illustrate the hedgehog model; the author says diversified service companies may grow and earn well but are unlikely to create something disruptive. Focus alone is not enough: people, strategy, and passion matter too. Patience helps avoid rushed, superficial work, while persistence lets teams revisit problems with new experience and ideas. The author also recognizes that crises demand intense effort, and describes balancing that urgency with patience as difficult. ### Source excerpt Great things take time. Evolution and complexity are two sides of the nature development. Let's take cars. Cars are becoming more and more complex. At the same time, cars are becoming more reliable, faster, more efficient, safer and better. It took humankind a century to get there. The progress pace is impressive, but incomparable with software development. Things change in software development world with an astonishing "never seen before" speed. 10 y.o. technologies are history. 10 y.o. software without major upgrades is dead. While 10 y.o. cars are quite good and decent. We, software developers, are blessed with a gift that makes our industry truly different -- we can upgrade working systems on the fly. To have a new car you should sell an old one and buy the new car. To have a new version of SaaS application you don't have to move a finger. Software developers have a luxury to iterate fast, fail fast and emerge viable, efficient solutions. Software development industry is not so young, it is about 50 years old. It took us 50 years to reach the point where "software is eating the world". Companies: Hedgehogs and Foxes OK, back to the main point of this essay. There are two types of companies: Hedgehogs and foxes. Hedgehog company takes one area and digs deep. It usually create a single software application for a single market (no matter if niche or wider). Fox company builds many applications for various markets, it constantly tries many new things and never stick to a single thing. I don't like foxes. In most cases they can't create something significant. However, let's check some examples. Rovio created 50 or so games till the Angry Birds slam-dunk It's a hedgehog company. It was fully focused on game development. They tried various models and various gameplays. They iterated and failed many times. Apple is a hedgehog company (well, there were different times at Apple, but now this statement is correct). Apple launches new products with care. No rush jumps to fin