From Liverpool’s Tide Marks to GPT-4 and 2026’s Smart Tech Picks
Britain’s Ordnance Survey has more than half a million recorded benchmarks, while sea level at Newlyn rose by an average of 1.83 mm per year from 1915 to 2015. Two very different source stories—James Bridle’s AI-assisted chair experiment and PCMag’s 2026 gift selections—offer a useful lens on how we measure, make, and use technology.
The Old Benchmark: A Shared Reference for Measuring the Land
Did you know that one of Britain’s foundational measurements began with a small chisel mark on a Liverpool church? In 1840, a surveyor cut a horizontal line and a broad arrow into an outer wall of St. John’s Church, on Old Haymarket. The mark indicated a height of 100 feet above sea level and gave surveyors a repeatable place to position a leveling rod. The device called a bench supported that rod; the mark and arrow together became known as a benchmark. James Bridle’s article, “Benchma(r)king,” traces how this practical mark became part of a country-wide system for comparing elevations. It is a story about measurement infrastructure rather than a new technology launch, but it has immediate relevance to cities: reliable maps and elevation references underpin how people understand the built environment and changes to it.
The Liverpool mark did not remain Britain’s definitive reference forever. Surveyors later chose mean tide level as a more accurate starting point, observing tidal fluctuations for nine days at Victoria Docks in Liverpool. That reference was eventually moved to Newlyn, Cornwall, in the 1920s. Its Ordnance Datum Newlyn, or ODN, provides the national elevation reference described in the article, while recognizing that “sea level” is not identical everywhere. At Newlyn’s tidal observatory, built on the South Pier in 1915, an inlet pipe lets the Atlantic tide fill and empty a hollow tower. Floats suspended in the tower connect to a recording device, turning water-level movement into measurements. This is not a satellite algorithm: it is a physical observation system whose readings establish a datum that can be used elsewhere.
The measurements also show why a reference point cannot be treated as timeless. The article reports that mean sea level at Newlyn rose by an average of 1.83 millimeters per year between 1915 and 2015, and by 3.8 millimeters per year between 1993 and 2014. Each change affects how elevations are understood relative to the datum. The last official benchmark was carved in 1993 on a milestone outside a pub in Loughton, Essex; satellite-based systems later took over the datum. Even so, the Ordnance Survey still maintains records for more than half a million benchmarks, though many marks have disappeared. These details complicate the familiar modern idea that digital positioning simply replaces older infrastructure. GPS antennas may now sit atop the Newlyn tide tower, but satellite measurement still exists alongside a history of physical marks, water observations, and reference systems.
Bridle asks readers to remember that the original purpose of a benchmark was not to rank one object against another. It was to provide a stable point from which measurements could be repeated and connected. That distinction matters when the word is now commonly used for judging software performance, product quality, profitability, or growth. In a city context, a benchmark can help make different measurements comparable—but only if people understand what is being measured, how the reference was established, and when it needs revision. The source does not claim that the Liverpool mark or ODN directly governs a particular urban project. As an editorial inference, however, the history offers a practical lesson for smart-city data: accurate-looking dashboards are only as useful as their underlying reference points. The next decade’s urban systems will need not just more sensors, but clear, maintainable ways to connect their readings to the places and conditions they describe.
James Bridle, GPT-4, and a Chair Designed as Collaboration
A second story begins not in a lab or a furniture factory, but on an apartment balcony. James Bridle—a writer, artist, and technologist—describes turning to carpentry while living in Athens during a difficult period for their family. Their partner, Navine, became severely ill at the end of 2023 and spent nine months in hospital; Bridle was also caring for their three-year-old son. With limited tools and saved scraps of wood, Bridle asked ChatGPT, then running GPT-4, to design a chair. The request was deliberately practical: provide measurements for a small collection of available materials and use the system’s ability to optimize a design, while leaving the hands-on work to a human maker. Bridle frames the experiment with the question, “What if we used AI for something actually useful?” The source does not present this as a product announcement or a controlled test; it is a personal account of trying an AI system as a workshop collaborator.
The process combined human-provided constraints with machine-generated design suggestions. Bridle measured scraps of plywood, pine panels, chipboard, dowel, and beam, then supplied the inventory to GPT-4. The tools included a saw, drill, screws, and wood glue. The system’s proposed role was not to manufacture the chair or take over every decision, but to help with calculations and offer ideas. Bridle describes computers as good at optimization: they can reduce a designed object to fewer or more efficient elements while maintaining its function, sometimes producing forms a person would not have imagined. In this case, the goal was a seat that could support an adult and remain comfortable, a deceptively demanding design challenge. The source does not provide the full final design specification, a load test, or a measured comparison against a human-designed chair, so it would be misleading to claim quantified performance gains.
Bridle’s approach draws on Enzo Mari’s “self-design” philosophy. Mari, an Italian designer, published Proposta per un’autoprogettazione in 1974, collecting nineteen furniture designs with photographs and simple instructions. The project encouraged people to make and adapt pieces themselves, using basic materials and carpentry. Mari’s instructions invited builders to send photographs of their variations back to his studio, treating design as an open process rather than a sealed expert product. His critique of status-driven design is captured in the source’s direct quotation: “If everybody likes it, it means I have confirmed the existing reality and this is precisely what I don’t want.” Bridle’s AI experiment echoes that emphasis on participation: the system can contribute calculations or options, while a person still cuts, assembles, judges, and lives with the result. The comparison is an interpretation, not a claim that Mari endorsed AI.
This small-scale experiment suggests a useful distinction for the AI industry: assistance is not the same as replacement. Bridle wanted a “smart carpenter’s mate,” not a machine that took away the rewarding parts of making. That framing has technical consequences. A system can provide candidate solutions, but the human still needs to check whether the plan fits the available material, tools, skill, safety requirements, and intended use. The story also recalls Bridle’s account of AI’s recent resurgence: they argue that expanded processing capacity, data storage, digitized material, and online activity helped supercharge existing approaches such as neural networks. They characterize the resulting concentration of power and data in a few companies as central to the current AI moment. For urban technology teams, the lesson is not that GPT-4 should design public infrastructure without oversight. Rather, as an editorial inference, generative tools may be most constructive when they help people explore options while leaving accountability, context, and physical verification with humans.
PCMag’s 2026 Gift Picks: Personal Devices, Specific Capabilities
PCMag’s first Gift of the Year awards select a dozen products its experts consider especially suitable for gifting, rather than declaring them the year’s overall best products. The article says affordability was part of the selection approach and includes devices across computing, phones, entertainment, health, reading, and hobbies. Among the named picks, Dell’s budget-focused XPS 13 is described as a Windows laptop priced below $800, with a colorful touchscreen, responsive keyboard, and thin, light chassis. Apple’s Mac mini with M6 is presented as a powerful desktop option below $1,000, despite a price increase associated in the article with a memory shortage. Google’s Pixel 11 is described as thinner and lighter than its predecessor, with a rear triple-camera system whose main sensor is 50% larger. Those are source-reported product descriptions and awards, not independent conclusions about what every buyer should choose.
The technical specifics help explain why the picks are differentiated. PCMag says the Mac mini’s M6 includes Apple’s most potent chip yet, an upgraded neural coprocessor, more advanced graphics, and faster memory bandwidth. For the Pixel 11, the larger main camera sensor is described as capturing more light, while the phone’s software includes Magic Capture, which extracts key photos from video, and Magic Cue, an on-screen assistant for contextual actions. The article reports 19 hours of battery life for the Pixel 11. These details describe distinct layers of a device: processor and memory for computing, camera hardware for gathering light, and software features for selecting or presenting information. The source does not explain the internal algorithms behind Magic Capture or Magic Cue, so their operation should not be embellished with unsupported claims about models or architectures.
The gift guide also includes products that connect personal technology to information and observation. The Vizio VQM-series Mini-LED Quantum TV supports Apple AirPlay and Google Cast, allowing streaming from phones, tablets, or computers; the 55-inch model is listed below $280 and the 85-inch model below $800. The Anker Soundcore Liberty 5 Pro earbuds offer active noise cancellation, multipoint Bluetooth for up to three devices, and a touchscreen charging case. For health tracking, the Apple Watch Series 12 is described as having a 54-hour battery and a sensing system that regularly checks heart rate and other vital signs during and between workouts. The Oura Ring 5 tracks activity, sleep, and stress, presents longer-term trends such as cardio capacity and cardiovascular age, and lasts more than 10 days per charge. These are particularly concrete examples of devices collecting or presenting personal data, though the article does not provide clinical validation details.
The list is a snapshot of consumer technology, not evidence of citywide adoption or a forecast of market share. PCMag names product reviewers and editors, including Joe Osborne, Kimberly Gedeon, Andrew Gebhart, and Will Greenwald, and labels several selections Editors’ Choice picks. It also includes an Onyx Boox Go 6 Gen 2 e-reader running Android, a Birdfy Feeder Rookie camera that records 30-frame-per-second video, and the 2.9-pound DwarfLab Dwarf 3 smart telescope, controlled entirely through a mobile device. For smart-city readers, these examples are relevant as design patterns—interoperability, sensing, long battery life, and user-facing interpretation—not as proof that consumer gadgets are equivalent to municipal systems. A hypothetical application might use these design principles in a public-facing service, but the source does not report such a deployment. The important question for the coming decade is how to make useful sensing understandable and trustworthy, whether the device sits on a finger, a bird feeder, or public infrastructure.
From Personal Sensors to Urban Measurement: What the Sources Do—and Don’t—Show
Together, the two articles offer a contrast between measurement as a shared foundation and technology as a personal tool. Bridle’s benchmark history describes an elevation reference built from physical marks, tide readings, and a maintained record. PCMag’s gift guide describes products that gather or process information for individual users: a ring tracking sleep and stress, a watch checking vital signs, a camera capturing birds, and a telescope producing images through a mobile-controlled device. These are not interchangeable systems. ODN is a geodetic datum, while a health tracker’s readings concern an individual wearer and a camera’s footage concerns the scene in front of it. The distinction matters because a sensor’s output becomes meaningful only in relation to its purpose, measurement method, and reference. Neither source claims that consumer devices are currently integrated into a municipal data network.
There is nevertheless a useful editorial connection for city technology. Urban teams routinely need to compare observations gathered in different places and at different times. The benchmark story shows how a physical reference can support repeatable measurements across a wide area; the products in PCMag’s list show how compact devices can gather, display, or transmit information. A hypothetical city service might draw inspiration from the consumer emphasis on clear interfaces and long battery life while still relying on separately validated sensors and defined measurement standards. That is speculation, not a reported deployment. It would also be a mistake to assume that more data automatically yields a better public decision. The source accounts instead point toward questions that are often less glamorous than a new device: What does the measurement mean? Can another person repeat it? What happens when a baseline changes? Who checks whether a generated suggestion or sensor reading fits the real situation?
Bridle’s chair experiment adds a third element: human judgment in the loop. GPT-4 was asked to work with a measured inventory of materials, but Bridle’s carpentry skills and practical decisions remained part of the process. Enzo Mari’s self-design project likewise emphasized making, adapting, and learning through construction. In a hypothetical smart-city setting, that collaborative pattern could inform how teams use AI to explore design alternatives or organize technical information: define the constraints, inspect the suggestions, and retain human responsibility for the final choice. The source does not establish that GPT-4 is safe or suitable for designing public infrastructure, nor does it supply a formal evaluation of the chair. Those limits are important. A prototype that is useful in a personal workshop is not evidence that the same workflow meets public-sector requirements for safety, accessibility, durability, or accountability.
For the next decade, the most promising theme here is not “more automation” but better relationships between measurement, tools, and people. ODN’s history illustrates the value of common reference points that can be updated when observations change. The gift guide highlights practical product capabilities—interoperability through AirPlay and Google Cast, multipoint connections, extended battery life, and interfaces that translate device activity into visible information. Bridle’s experiment asks whether an AI system can contribute without claiming the whole task. As an editorial conclusion, smart-city systems will earn confidence when their outputs are legible, their reference points are explicit, and people can question or verify what a machine proposes. That combination could make technology more useful in streets and public services, while keeping the human experience—not a benchmark score or an automated recommendation—at the center.
Where Robles.AI Fits
The sources do not describe Robles.AI’s products, services, deployments, or technical capabilities, so it would be inaccurate to claim that the company provides a particular platform or could have prevented a problem. The specific areas discussed here nevertheless identify a clear kind of capability that would be relevant to a company working in AI: evaluation of AI-generated recommendations, human-in-the-loop workflow design, and the disciplined use of reference data. In Bridle’s example, GPT-4 receives a defined set of wood dimensions and contributes design suggestions; a human checks those suggestions against materials, tools, and use. Any Robles.AI offering in this area would need to be verified independently. The article itself supports the broader need for collaboration and scrutiny, not a claim about what Robles.AI already sells.
A second concrete connection is benchmark and sensor-data quality. The Newlyn account describes how tidal observations become a reference datum, how that reference changed over time, and how the Ordnance Survey continues to maintain records of more than half a million benchmarks. For a city-focused AI service, relevant work could include documenting data provenance, checking whether measurements use compatible references, and making uncertainty or changes visible to users. These are proposed areas of relevance, not assertions that Robles.AI performs them. Similarly, the consumer devices described by PCMag—such as the Apple Watch Series 12’s vital-sign sensing or the Oura Ring 5’s long-term health trends—raise questions about what data a system summarizes and how a person interprets it. The source gives product features, not evidence of municipal integration or a specific data-sharing model.
An appropriately grounded Robles.AI connection, then, is a set of possible evaluation questions rather than a product endorsement: Can an AI suggestion be traced to its inputs? Can a human review it? Are the limits of a sensor or model communicated? Is the reference against which a result is compared still appropriate? Those questions follow directly from the chair experiment’s human collaboration, the benchmark history’s changing datum, and the gift guide’s examples of personal sensing. If Robles.AI offers services in AI evaluation, data quality, or human-centered system design, those would be the specific capabilities to examine against these cases; the provided sources do not confirm that it does. That distinction keeps the discussion honest while showing where expertise in responsible AI could matter.
Looking toward the next decade, cities will continue to encounter both physical measurements and machine-generated suggestions. A useful system will need more than an impressive interface: it will need understandable inputs, repeatable methods, and clear responsibility for decisions. Bridle’s “smart carpenter’s mate” is a compact image of that balance: computation can widen the set of options, while people remain responsible for the craft and consequences. Newlyn’s tide record adds the complementary lesson that even a trusted baseline may need revision as measurements change. Robles.AI’s place in this conversation can only be established by examining its actual offerings, but the test is specific: can its work help make AI outputs inspectable and data references dependable, without presenting automation as a substitute for human judgment?