Posts

Showing posts from October, 2026

Before I Travel: My Plan for a Lost iPhone, Not Just a Flat Battery

Image
A recovery plan belongs on the packing checklist too. AI-generated illustration; not a photograph of an actual loss or theft. I spend plenty of time thinking about what keeps my phone working on a business trip: a reliable charger, the right cable and enough battery to get through the day. But a fully charged phone is not much help if it disappears from my pocket. Since my iPhone is my main phone, losing it would interrupt much more than my photography. Tickets, messages, maps and access to accounts can all depend on that one device. That makes a recovery plan part of my travel kit, even if it takes up no space in the bag. Here is the checklist I want ready before leaving, with one distinction at its centre: erasing a stolen iPhone and removing it from Find My do very different things. 1. Make the security decisions before the airport First, check that Find My is enabled. These protections need to be in place before the phone goes missing; they are not something to add afterwards....

Instagram’s 90-Minute Videos: More Reach, or Another Unpaid Afternoon?

Image
A longer upload allowance does not create more hours in the day. AI-generated illustration; not an Instagram interface or a photograph of my workspace. My first reaction to Instagram testing videos up to 90 minutes long was not to start planning a 90-minute video. It was to think about the extra editing, uploading and checking that another format could add to a small creator’s week. I like the possibility of giving a topic more room. A travel story, a conversation about technology or a careful explanation of a spending decision can need more than a short clip. But a new place to publish is only useful if the work fits around everything else. A limited test, not a new requirement for everyone Instagram is testing a format called Features for horizontal videos up to 90 minutes. According to TheWrap’s reporting , the test began with certain creators in Brazil and is expanding to selected US creators. The accompanying Clips as Reels feature is intended to connect short extracts with ...

Apple Creator Studio: Can the New Tools Replace a Subscription I Already Pay For?

Image
The value is in the work a tool helps finish. AI-generated workflow illustration; the screens are conceptual, not Apple Creator Studio screenshots. A creative subscription becomes interesting to me when it can take something off my monthly bill. A longer feature list, by itself, does not do that. That is how I am looking at the latest Apple Creator Studio updates. There are useful ideas here for editing video, working on images and preparing documents. But the question for Routes, Bytes & Budgets is whether those tools fit work I already do, on devices I already own. This is my assessment of the announced changes before their release, rather than a hands-on review of the new features. The interesting part is the work between apps Apple has scheduled the updates for November, included for existing subscribers. Some Freeform additions follow in December. The changes that catch my attention are built-in MotionVFX content and stock video in Final Cut Pro, image editing that moves...

AI Chips Can Arrive Before the Power. The “Half Unused” Claim Needs Care.

Image
Around 20% of planned global data-centre capacity could face grid delays by 2030, in the IEA’s 2025 analysis. This does not measure GPUs currently idle. Graphic: Routes, Bytes & Budgets, based on IEA, Energy and AI (2025) , CC BY 4.0 . Redrawn as a percentage bar. A graphics processor can be ordered, delivered and paid for without doing a single useful task. Somewhere between the purchase announcement and the working AI service, it still needs a building, cooling, networking and a dependable electricity connection. That gap is becoming one of the most interesting parts of the AI boom for me. It also needs a more careful explanation than the claim that half the industry's chips are simply gathering dust in boxes. Where the “half unused” argument comes from In his September 22 analysis, Ed Zitron argues that more than half of Nvidia's AI GPUs have yet to become operational capacity . He builds that case from spending, infrastructure disclosures and estimates of installed ...

Alibaba’s Pitch for Europe: Can Open Models Deliver AI Independence?

Image
AI-generated illustration for Routes, Bytes & Budgets: an organisation operating local computing infrastructure. This is a conceptual scene, not an Alibaba office or a tested installation. I like the idea of software that remains useful even if its supplier changes direction. It is one reason open models interest me: keeping a working copy can offer a different kind of control from depending entirely on a remote service. That is the appealing part of Alibaba chairman Joe Tsai's argument about Europe's AI future. But I think the practical meaning of “sovereignty” deserves more attention than the slogan. Tsai's proposal: use open models and keep control locally Speaking at Wave by Vento in Turin on October 7, Tsai argued that open-source AI offered Europe its route to technological independence. In the event coverage , he describes European companies adapting models using their own industrial data and running them on infrastructure they control. I understand the attr...

Microsoft Is Rebuilding Around AI Agents. I’m Watching the Bill.

Image
Satya Nadella in 2017. Official Microsoft portrait by Brian Smale / Microsoft , via Wikimedia Commons, CC BY-SA 4.0 . Commons resized rendition; no further edits. When a company starts organising itself around AI agents, I pay more attention than when it adds another chatbot button. Microsoft is making that bigger move. The question for me is what happens to the work, the permissions and the bill once software starts carrying out tasks on our behalf. There is also a spending figure that needs updating. The widely repeated $190 billion was a forecast, not an amount Microsoft had already spent on AI. Its later guidance gives us a more useful starting point. The $190 billion headline needs a date In its July 29 earnings call , Microsoft put expected calendar-year 2026 capital expenditure at approximately $175 billion . The company said its underlying investment expectations were unchanged. The difference comes from accounting: revised useful-life assumptions for data centres and offi...

Anthropic Opens Its IPO Books. The $80.7 Billion Loss Needs Context.

Image
Dario Amodei at Downing Street in May 2023, shown in an archive photograph. Photo: Simon Walker / No 10 Downing Street, via Wikimedia Commons , CC BY 2.0 . Existing Commons crop; no further edits. I find an AI company's accounts more revealing than another leaderboard victory. Benchmarks tell us something about a model. Financial statements help explain the business that has to keep serving it, maintaining it and paying for the infrastructure behind it. Anthropic has now opened an important window into that business. The numbers are striking, but the largest loss figure is also the easiest one to misunderstand. A public filing, with the offering still ahead In its October 26 announcement , Anthropic said it had filed an S-1 registration statement for a proposed initial public offering. It intends to list Class A shares on the Nasdaq Global Select Market under ANT . The number of shares and price range were not yet determined, and the registration statement was not yet effectiv...

RTX 5090s Are Being Rebuilt for AI. That Doesn’t Make One a Better Buy.

Image
A standard Palit RTX 5090 GameRock, shown for product identification; this is not one of the reported modified boards. Photo: PantheraLeo1359531 , CC BY 4.0 . Commons resized rendition. A flagship graphics card can leave the factory ready for games and end up being dismantled before it renders a single one. That is the striking part of reports about RTX 5090 cards being converted for AI servers in China. My first reaction is that this says something about how valuable the hardware inside a gaming product has become. My second is more practical: it is a good reason to understand the market, but a poor reason to panic-buy a graphics card. What the reports actually show Tom's Hardware reports footage of Chinese workshops removing GPU and memory components from RTX 5090 cards, then fitting them to custom boards with blower-style cooling for server use. The reported process explains how consumer hardware can be repackaged. It does not establish a reliable industry-wide count of co...

Musk Warns of a Neocloud Shakeout. I Want an Exit Plan for My Data.

Image
AI-generated illustration for Routes, Bytes & Budgets. This conceptual server room does not depict a particular cloud provider. The phrase “cloud computing” makes everything sound light and easy to move. Behind it are buildings, power contracts, expensive processors and financing that still needs paying when a customer stops renting. That is why Elon Musk's warning about specialist AI cloud providers caught my attention. I am less interested in picking the next company to fail than in what the argument means for anyone choosing where to run a project. A warning from someone with a stake in the market According to Reuters' October 26 report , Musk predicted a wave of failures and consolidation among the more than 200 companies he described as neoclouds. He also said SpaceX would not buy them. This is a prediction, not evidence that every provider is in trouble. Musk is involved in this market himself. On October 26, SpaceX announced a binding commitment for approximatel...

How to Check Your Website’s PHP Patch Status—and Read the Result Properly

Image
AI-generated editorial illustration of a website maintenance check. The screens do not show an actual PHP scan result.   When a serious web vulnerability is announced, the first question is usually straightforward: does this affect my site? The answer can be frustratingly difficult when a hosting dashboard, a terminal and a website header all reveal different pieces of information. OpenAI has released a small PHP version checker to help with the two critical PHP vulnerabilities disclosed on October 21. I see it as a useful starting point for investigation. Its output needs interpreting carefully. 1. Find out who manages the server If you run your own PHP application or self-hosted website, identify the person or provider responsible for the runtime. If you use a managed publishing service, begin with that service's status and support information. A Blogger author does not have a PHP runtime to upgrade through the post editor. For managed PHP hosting, a clear support request is oft...

PHP’s Security Crisis Shows Who Really Keeps the Web Running

Image
AI-generated editorial illustration of software infrastructure maintenance; not a photograph of a compromised server. We spend a lot of time discussing the newest AI model. Much less attention goes to the established software that keeps websites working every day—until something serious breaks. The October PHP security releases are a reminder that maintaining that software is demanding work. They also make the relationship between AI and open source more complicated than “the machines will find the bugs for us.” What happened to PHP On October 21, the PHP project released fixes for two critical vulnerabilities, CVE-2026-109618 and CVE-2026-109619, that can allow remote code execution under applicable conditions. Its official advisory calls for immediate updates across PHP installations serving web requests. The important distinction is that these are problems in PHP itself. Updating a website's theme or plugin does not establish that the underlying runtime has been patched. The ho...

Unitree Wants Robots to Improve Themselves. What Would Count as Proof?

Image
Unitree robots at Japan Mobility Show 2025 in Tokyo. This is an earlier hardware demonstration, not a Rehearsal test. Photo: RuinDig/Yuki Uchida, Wikimedia Commons, CC BY 4.0. Unaltered. Image source · CC BY 4.0 licence . A robot doing a backflip is hard to ignore. A robot becoming more useful without someone manually programming every improvement might be the more important development. That is why Unitree's latest announcement caught my attention. The company is working on a system called Unitree Rehearsal, intended to automate more of the process of developing and improving robot behaviour. The interesting question is how well that process transfers from a computer simulation to the physical world. What Unitree is proposing Speaking at a physical AI conference in Hangzhou on October 22, founder Wang Xingxing described a development loop in which AI would search research papers, write code, test it in simulation and move on to trials with physical robots. The account of his anno...

Google’s AI Search Got a Doctor Wrong. A German Court Stepped In

Image
AI-generated editorial illustration of source checking; not a screenshot of Google’s disputed answer or a person involved in the case. I like getting a quick explanation before opening a pile of tabs. What worries me is how easily a fluent AI summary can make a weakly supported claim look settled—especially when it is about a real person. A German court case involving Google's AI search brings that problem into focus. It is about responsibility for a harmful statement, and about what a source link does—and does not—prove. The Aachen case, without exaggerating its scope According to Höcker, the law firm representing the claimant , Google's AI Mode attached serious false criminal allegations to a doctor by confusing him with another person. The linked reporting did not name that doctor. The Aachen Regional Court issued an interim injunction on September 24, 2026, case 8 O 268/26, prohibiting the disputed statements. Legal publisher beck-aktuell's account confirms the case de...

IBM Says Open-Weight AI Costs Less. I’m Counting the Whole Job

Image
AI-generated editorial illustration of AI infrastructure budgeting; not an IBM product photograph or facility. I am always interested when the AI conversation moves from “which model is the most impressive?” to “what does it cost to get useful work done?” Those questions can produce very different shopping lists. IBM chief executive Arvind Krishna has made a striking argument for open-weight models. During the company's October 21 earnings call , he put their effectiveness at roughly 90% of frontier alternatives, with token costs around 10–25% as much. That is Krishna's assessment of the market, not a universal benchmark result or a promise that every organisation can cut its AI bill by 90%. The range matters, and so does what sits outside the token bill. “Good enough” depends on the job A model that is excellent at categorising routine enquiries may be less convincing when asked to resolve a difficult technical problem. Averaging those abilities into one number hides the decis...