tech

2024-09-01 · open the original ↗

• Classify mixed documents first, then run the matching extraction model. Giard's reason: invoices, order forms, agendas, or invoices from different companies put fields like the company name in different places. Classifying a document first tells you which layout-specific custom extraction model to use. He covered that model in episode 176.
• A classification project needs an Azure AI service and blob storage. In Document Intelligence Studio (documentintelligence.ai.azure.com, which redirects to /studio), he picks "custom classification model" and creates a project. He creates a new resource group, a new AI service (formerly called Cognitive Services) in East US on the standard pricing tier with the default API version, and a new storage account and blob container. Storage account names must be all lowercase, and he took the cheapest tier. Setup took about 55 seconds, mostly spent creating the storage account.
• Each document type needs at least five labelled samples. Files can be dragged into the studio or put straight into the blob container. Giard used five each of agendas, invoices and memos, each type sharing one layout. A type can be created with "add a type" or while labelling a document, which creates and assigns it in one step. A check mark shows when a type has enough samples to train.
• Training is quick, but the models list must be refreshed by hand. Training started in a few seconds. Giard says it can take about a minute depending on how many documents there are. The status read "not started" until he clicked refresh, then went to running and then succeeded.
• In the test tab, the model classified one document of each type correctly. One agenda, one invoice and one memo were each recognised as the right type. Giard says an API exposes the same classification to applications. His suggested flow is to put documents in blob storage, classify them, then run the custom extraction model for that type.

2024-08-30 · open the original ↗

• Marsman's Office of the CTO job is technology transfer and demos. At Microsoft she explores new OpenAI models as they arrive and works out which product groups and field staff need to know about them. She also helps Fortune 100 strategic partners, among them Etsy, whose CTO and search engineering head presented with her at Build 2024, and she is Kevin Scott's official demo engineer. She likens the role to a grad student serving a visionary professor: running experiments toward his vision and bringing insightful papers back to him.
• Hands-on, crowd-sourced prompting taught her DALL-E's limits within a month. She joined the office about two years before the recording, as DALL-E 2 released, and ran "30 days of DALL-E" on prompts solicited on Twitter; by the end, she says, she could tell from a prompt whether the model would handle it well. A later C# game on her GitHub shows an image and has the player guess its prompt. It grew out of coaching a colleague to add detail (viewpoint, empty or full field, a 1920s–30s look) to a bare "baseball stadium" prompt, and she says it is still wired to DALL-E 2.
• Chaining model calls produced a genuinely useful end-to-end copilot demo. For Scott's Build 2023 keynote she automated the social posts he hates writing for his podcast, Behind the Tech. Audio went through OpenAI's Whisper for speech-to-text, a local model extracted the guest's name, a Bing web search fetched a bio, and GPT-4 wrote the post plus a prompt for a DALL-E image; a tool written with the LinkedIn team then posted it. She presents it as the power of orchestration layers such as Semantic Kernel and LangChain.
• Small local models enable uses too costly for cloud LLMs. Her Phi-3 vision demo puts local recycling rules in the prompt and says whether an item held up to the camera can be recycled. She ran it on a laptop GPU and tested it offline on her flight to Build with a plastic cup, an Aquafina bottle and an aerosol can; it was cut from Build 2024 for time but is on her GitHub. Her argument is that nobody would pay GPT-4-class prices at the scale of public-park bins or a recycling center, while a local model removes the cost, latency and network dependence.
• Her boss pre-empted her worry that a staged mistake looked bad. She got access to GPT-4o's audio-in, audio-out mode only at 1 PM the day before the Build 2024 keynote, and built a demo of asking it for help with a simple Python error. She feared that erring deliberately in public reflected badly on women in tech. She says Scott, unprompted and without his comms staff in the room, proposed saying on stage that she would never make that mistake in real life, and she gives this as the reason she trusts him.

2024-08-29 · open the original ↗

• One power bank can run a Starlink Mini and UniFi network. The host powers a Starlink Mini from a 27,600 mAh Anker Prime power bank through a USB-C-to-barrel cable, with Ethernet from the dish into a UniFi Express; he expects the battery to keep the whole setup online for about 3 to 4 hours.
• The UniFi Express serves as both gateway and access point. Because it has a built-in access point, the host says the single unit gives a full UniFi setup, with room to add switches or more access points.

2024-08-29 · open the original ↗

• The new Nature of Code is a full rewrite, not a refresh. Shiffman says the class began at NYU's ITP around 2008, and he self-published the book through an early Kickstarter, then left it largely unchanged for about 10 years. He brought in No Starch Press to "project manage" him and produce a better print object. The new edition is in JavaScript with p5.js instead of Java and runs to about 640 pages.
• Neuroevolution is the book's new centerpiece. The 2012 edition only covered a single perceptron. He rewrote that chapter around ml5, an open-source machine-learning library for p5 that he also works on, and added a chapter on evolving the weights of many random networks instead of using backpropagation. He calls it a lightweight way to play with machine learning in the browser with no dependencies; Hanselman notes the book builds from a dot moving at random to a Flappy Bird that learns to flap.
• Everyone should get the chance to code; not everyone must. Shiffman argues that understanding code is empowering when so much of life runs on software. He says teaching through art and creative exploration, with "no answer to this problem", brings in people who thought coding wasn't for them. He also says the book is not a learn-to-code book: it assumes variables, loops, conditionals, arrays, functions and objects.
• Giving the book away free sells it rather than cannibalizing it. The full text stays online under Creative Commons, and he calls the website marketing that widens the book's reach, while admitting he has only anecdotal evidence. He describes the free material as subsidized by his full-time NYU teaching job, since he counts open education materials as part of that work.
• p5.js reimagined Processing for the web, community first. Shiffman credits Casey Reas and Ben Fry with starting Processing in 2001, building on John Maeda's Design by Numbers, and calls its play button for running code revolutionary. As Java applets died, Lauren McCarthy rebuilt it for the browser and broke down the line between users and developers. The p5.js web editor hosts every example so code can be changed and shared from any machine, and he invites readers to port the examples to Processing, openFrameworks, Unity and other environments.

2024-08-29 · open the original ↗

• Treat Copilot as a rubber duck, not a one-shot oracle. While upgrading the Hanselminutes site to .NET 8, Hanselman says most people ask one question and stop if they dislike the answer. He builds up context instead with follow-ups ("tell me more about the setup method", "is this modern idiomatic .NET"), and keeps separate named chat threads for each piece of work.
• Scope what Copilot sees. He shows slash commands such as /explain working on selected code, and inline chat marking the exact scope it is reasoning over. He then widens that scope to a file plus its tests to ask whether the tests are affected.
• Generated commit messages are the standout feature. He expected the generated message to say only "added new tests". Having read the diff, it also listed earlier Dockerfile and .NET hosting changes he had forgotten. He calls this "the killer feature" and says he would have missed those changes himself.
• AI tools should show where they looked. In his demo, analyze-and-fix on a failing test immediately found the deliberately wrong assertion (690 expected, 694 actual), and ask-Copilot on a null-reference exception pointed to the line he had broken. He praises it for listing the references it used and argues no AI app should be a black box, so a user can see why it answered and correct it.
• The small "delighters" go unnoticed, and AI icons are inconsistent. He shows rename suggestions and conditional-breakpoint suggestions (it proposed "shows greater than 10" from context). He complains that the AI is marked with a sparkle in one place, a sparkle pen in another and a Copilot figure in a third, and says the industry hasn't settled on one icon. He recommends Help > What's New as the way to find such features.

2024-08-27 · open the original ↗

• Meshtastic gives cheap, low-power messaging where cellular fails. The host pitches the open-source mesh for events with congested cell coverage (a music festival, the Baja 500), remote hiking with a base camp, farm IoT sensors on small solar panels and batteries, and GPS asset tracking on vehicles, before handing off to Crosstalk's David Barger for the build.
• The recommended RAK Wireless starter kit needs a few add-ons. Barger bought three at roughly $20–30 each: a RAK19007 baseboard with a 900 MHz LoRa module (US), an optional GPS module he calls power-hungry, and a screen. The battery (about $2–3) and enclosure are extra; his battery connector did not fit and had to be re-soldered. Connect antennas before powering on, or risk frying the board.
• Set the LoRa region first, or the node will not work. Barger flashes the latest stable firmware via the web flasher in a Chromium-based browser, then sets the region from "unset" to US. Defaults to know: the primary channel has no pre-shared key, so add one on a secondary channel for privacy; roles are client (relays), client mute (never relays), router and repeater; leave the LongFast modem preset as a beginner.
• One high node covered about seven to eight city blocks. With a large 915 MHz antenna on a town's tallest parking garage, Barger reached the fringe at seven or eight blocks, shorter where large buildings intervened. To cover the downtown he would place repeaters about six to seven blocks apart, as high as possible.
• The firmware is still beta, and it showed. Nodes left on overnight stopped receiving each other's messages despite transmitting; after hours with the logs, only a factory reset fixed it, cause unknown. His GPS module never got a lock even outdoors, possibly specific to his hardware; a phone can share its GPS instead, which he says is no good for standalone asset tracking.

2024-08-26 · open the original ↗

• Aim the Starlink Mini roughly north, then fine-tune alignment. The host uses a compass to point the dish in the general direction of north, plugs it into the RV's power, and corrects the alignment with the app's guide; the obstruction view showed trees overhead.
• Speed roughly sextupled once the dish settled in. The first test, with obstructions, gave about 22 Mbps down and 10 up, already above the campground's paid Wi-Fi, which tops out at 10 Mbps. After 20 to 30 minutes of settling, the host measured 128 Mbps down and 20 up.

2024-08-22 · open the original ↗

• A technical product chief reads customers better. Frederick, chief product officer at Stanza Systems, says engineering experience lets her tell whether a pain point is coming from product, platform engineering or SRE staff detached from business goals. She says a CPO without it is at best slowed down and more likely hindered.
• CI/CD maturity should fit the business, not be chased. She cites a team that pushed for faster releases while its heavy manual testing burden made each release more costly. She argues that an early-stage product still finding its business may not repay full test automation, and Hanselman concedes this challenges his instinct to automate anything done twice.
• Stanza moved its insights product ahead of schedule. Frederick says customers confirmed the problem behind Stanza's first product, rate limiting, but showed little urgency in a cost-focused market. They had also already paid heavily for observability and would not add another tool. So Stanza decided to draw insights from data customers already collect, a step it had planned for later.
• Tooling sped up building software but not running it. She argues that small teams can now produce far more software than they can operate, and that this leads to burnout. On "shift left", she agrees with Hanselman that shrinking budgets often just dump right-side work (performance, reliability, resilience) onto developers, "borrowing" development time.
• Stanza wants to codify senior engineers' judgment. She frames operations as signals, comprehension and actions, and says signals and actions are well served while comprehension still relies on senior people sifting the data. Stanza aims to systematize that triage into prioritized clues, such as flagging a service that has become risky. Hanselman adds that this should make junior engineers better faster without letting them stop thinking, and she agrees juniors learn most by watching how seniors reason.

2024-08-16 · open the original ↗

• Password Pusher replaces clear-text passwords with expiring links. The sender sets expiry by days or views (the demo uses four days or three views), optionally adds a retrieval passphrase, and shares a link or QR code. The host argues the hosted site is low-risk because only the password is sent, not the username or site, and the code is open source; self-hosting removes the trust question.
• A $5-a-month cloud server is enough, with backups off. The host uses a Vultr shared-CPU instance with the one-click Docker image (25 GB SSD, 1 vCPU, 1 GB RAM; $1 a month more for AMD high-performance) and disables backups, because the app stores nothing worth restoring. He creates the DNS A record early, in Cloudflare with proxying off, to give propagation time.
• Install is two Docker containers and one Caddyfile edit. The host downloads a Docker Compose file and a Caddyfile, sets the Let's Encrypt email and the fully qualified domain name in the Caddyfile, and runs docker compose up -d, which pulls the Password Pusher and certificate containers. He says the whole job takes under 15 minutes; written steps are on the Crosstalk blog.
• Lock SSH to your WAN IP and auto-restart the containers. With ufw, he allows port 22 only from his own WAN IP (a /32 for a single address), allows HTTPS 443 from anywhere, deletes the default anywhere rules for 22, and enables the firewall. Adding restart: always to both services in the Compose file brought the site back within a minute of a reboot.
• Crosstalk itself uses Password Pusher with its customers. The host says the firm sends and receives customer passwords this way, and presents it as showing customers it values security on their networks and phone systems.

2024-08-15 · open the original ↗

• Kubernetes' declarative state made bespoke deployments obsolete. Garfield says Codefresh bet early that containers would change everything and that Kubernetes would win. Declarative state management, he argues, turned hand-rolled, special-case deployments into something repeatable and automatable.
• GitOps was formalized to stop it meaning everything. He relays that the term was coined at Weaveworks by Alexis Richardson after git-stored state allowed a fast recovery from an outage. A CNCF working group started around late 2020 with Codefresh, Azure, GitHub, Amazon and Weaveworks, then Red Hat, drawing in about 80 companies and 120-plus parties. It reached 1.0 of four principles by about 2021: git as the source of truth, plus a reconciliation engine that monitors live state for drift.
• Argo CD has effectively won Kubernetes deployment. Garfield estimates that a little over half of Kubernetes deployments use Argo CD, against about 10% for the runner-up. He says the figure is extrapolated from a CNCF survey and is not exact. Codefresh launched the first commercial Argo CD around 2020 and calls itself the top security contributor to the project.
• Octopus Deploy bought Codefresh for cloud-native reach and funds its open source. The acquisition closed in March after the Octopus founder met the team at KubeCon. Garfield says Octopus is strong on VMs and Windows/.NET deployments, where Codefresh was weak. Octopus committed to expanding the open-source team, and he has been hiring additional Argo CD maintainers.
• Kubernetes at the edge is everywhere and cheap. Garfield names Starbucks and Chick-fil-A stores, F-16s and container ships, and argues a two- or three-node cluster survives node failures. With GitOps, a disconnected cluster picks up its updates when it reconnects. He compares Kubernetes to snowboarding: painful on day one, then easier to master than the alternative. He warns that the power to bootstrap everything from git is also the power to destroy everything in one command, a topic for ArgoCon North America (12 November, Salt Lake City, $200 academic price).

2024-08-15 · open the original ↗

• Typing the bare command shows whether the SDK is installed. Giard runs dotnet in a terminal: if it prints help options the SDK is present; if the command is not recognized, the installer is at dotnet.microsoft.com/download.
• A new project takes its name from the folder unless overridden. dotnet new console in a folder called ConsoleTest1 scaffolded a one-line Hello World program as ConsoleTest1.csproj; adding --name (he used "GCast Console") produced a project with that name instead, otherwise identical.
• Build, clean and run are separate commands; run builds implicitly. dotnet build compiles and dotnet run runs the project in the current folder, building first if it is unbuilt or changed since the last build. Giard says dotnet clean is for when leftover artifacts, such as an old NuGet package after references are added or removed, confuse the compiler; it makes the next build slower but clean.
• Templates go beyond console apps, and C# is only the default. dotnet new mvc scaffolded a web app with models, views and controllers, which dotnet run served at localhost:5069 with a home and a privacy page; an edit to the home view showed on refresh. Adding a language argument set to F#, with --name, produced the same MVC project written in F#.
• The CLI documents itself, with fuller reference online. dotnet help lists every command (he points to dotnet test, which runs a solution's automated tests), dotnet help test opens that command's switches and arguments, and a page on learn.microsoft.com covers all of the commands.

2024-08-08 · open the original ↗

• Tech creators already do the work and deserve pay. Freeman, co-founder of the influencer-marketing firm Freeman & Forrest, says influencer marketing is huge in fashion and beauty but has not landed in tech. She argues that creators already raving about products should be paid, just as the industry accepted paying open-source maintainers. The condition is authenticity, summed up in a line she credits to Corey Quinn: "you can buy my attention, you can't buy my opinion."
• Measure impressions, engagement and conversion according to the goal. Her firm measures all three, then hands converted users to the client's own funnel. A cash-limited startup needing signups gets a conversion focus, while trust-building campaigns weight engagement. She avoids leading with Twitter's "empty calorie impressions" and favours smaller, high-trust channels such as Cassidy Williams' newsletter.
• Paid creators extend developer relations' reach. From her developer relations work at Microsoft and AWS, she says in-house advocates exhaust their reach quickly and see returns collapse when they repeat themselves. She adds that she always resisted companies' requests to list influencers to target, and that payment lets "tinkerers" afford to keep experimenting.
• Lasting careers need substance, not just shine. Hanselman complains about engagement-bait influencers who ask generic JavaScript questions. Freeman says questions are fine and the bigger problem is gratuitous meanness, which she calls a five-year run at most. She describes visible tech people on a spectrum from shine (communication, authenticity) to substance (real coding experience) and says going too far toward either end causes problems.
• Today's newcomers can't replay the golden-age paths. Freeman says she entered code school and developer relations in their golden ages, and that people now trying to replicate that era misread a changed world. She credits years of "spinning in mud" for her grit. Hanselman, who admits his own career has been lucky, advises going one level below the abstraction you know, for example JavaScript beneath React. She endorses this as advice that at worst does no harm.

2024-08-06 · open the original ↗

• The host took nearly five minutes to terminate a Cat6 cable. Timing himself through cutting, stripping, removing the centre spline, untwisting and ordering the pairs to T568B, he finished in 4 minutes 54 seconds with a passing test, and says as a hobbyist rather than a professional installer he expects many viewers are faster.

2024-08-02 · open the original ↗

• Voice is the friction-free interface, and the obstacle is human, not technical. Russell, chief AI officer at the AI Leadership Institute, says she saw this when her son with Down syndrome and her father with a traumatic brain injury began talking to devices. She argues that letting users talk to your tech means they never return to your website. She says companies are slow to exploit voice because teams protect what they built, not because the technology is unready.
• Alexa is primed for a generative-AI reboot. Hanselman calls the device a dumb terminal. Russell says her 100-plus Alexa skills reach about 2 million unique users with unchanged code, such as an 800-line daily-affirmation skill that walks through 365 entries. She predicted the new generative Alexa was "weeks away" and said it would let skills run in 40 languages without extra developer work. She also says older (75-plus) and younger (under-20) users tolerate voice errors far more than people in between, and that generative AI lets devices fail gracefully.
• Voice AI's business case runs through loyalty, not direct revenue. Hanselman asks how a $59 device can sustainably call an expensive model. Russell points to Alexa team layoffs and to Bezos moving on from his Star Trek computer project, and says Alexa took about 10 years to get barely into the black. She says Alexa owners buy more on Amazon in general, not through the device, and argues builders need values that sustain them through costly retraining.
• Anthropomorphized wake words helped adoption but carried costs. Hanselman dislikes "yelling at invisible women" and welcomes unnamed products such as ChatGPT and Copilot. Russell argues the human name probably helped adoption, says "Alexa" was picked to minimize false wakes, and notes girls named Alexa were hurt, with mothers attempting a class action. She sees a return to push-to-talk in ChatGPT voice and the Humane AI pin.
• Disclose AI and recording up front, or lose more trust. Russell cites a Stanford study finding that using AI always lowers trust, but far more when people discover it afterwards. She says no legislation currently protects a person's ownership of their likeness, and criticizes a realtor-targeted voice product (captioned as "air") that never discloses it isn't human. Her prescription is to build responsibly even if it is slower, and to use AI red teaming with diverse people who judge experiences, not only adversarial attacks.

2024-08-02 · open the original ↗

• The WLAN Pi R4 kit turns a spare Pi 4 into a Wi-Fi toolkit. The host bought the bring-your-own-Pi kit himself, about $180 USD with shipping, and says the video is unsponsored with no contact with the project. It adds a heatsink case with display and a USB Wi-Fi 6 adapter, since the Pi 4's onboard radio is Wi-Fi 5. It needs a Pi 4 with 2 GB or more and did not yet support the Pi 5.
• Left at a client site, it becomes a remote Wi-Fi probe. Reachable over VPN, Tailscale, ZeroTier or Cloudflare tunnels, it can serve as a remote endpoint for WiFi Scanner on Windows or WiFi Explorer Pro on Mac. The host would pair this with its bundled Grafana dashboards, which log periodic speed tests, latency and DNS resolution time.
• The profiler shows what a misbehaving client can actually do. It broadcasts a test SSID; any device attempting to join, with any password, gets profiled for supported bands, beamforming and maximum transmit power. The host uses this to diagnose a client with a weak antenna or one that cannot use the available band.
• Kismet and Cockpit come bundled, but Kismet runs hot. Kismet, which the host likens to Wireshark for Wi-Fi, scans selectable channels and details every device and access point, but loads the Pi enough that the metal case matters. Cockpit gives a browser console with updates and a terminal; spectrum-analyzer dashboards exist for tools he does not own.
• Most functions work from the device's own screen and buttons. Without the web GUI, it offers reachability checks, speed tests, tests to vendor clouds (Aruba Central, Extreme, Mist, Ruckus), a port blinker for finding its switch port, a hotspot bridged to Ethernet, and a DHCP/TFTP server mode. The host had not tried its Bluetooth features.

2024-08-02 · open the original ↗

• Container queries style components by their container, not the viewport. Riviere calls them the next step in responsive design, now termed "intrinsic design", and suited to component-based front ends built with React or Vue. He believes they now work in all browsers, with Firefox the last to catch up.
• One container query can replace several media-query breakpoints. In his live demo a pagination div is declared container: pagination / inline-size, and @container pagination (max-width: 200px) hides the list of page links and shows a select dropdown instead. Placed in a fixed 200-pixel sidebar, the same component always shows the dropdown. On a Shopify cart sidebar he says an icon grid that moved between one and two columns would have taken four media queries but needed one container query.
• CSS nesting now works natively in the browser. Riviere says the nesting of styles and media queries that once needed a preprocessor such as Sass or Less is now plain CSS. He nests the container query inside the .pagination rule.
• CSS is strongly typed, and @property lets you declare types. His example is that CSS ignores a pixel value assigned to a colour. With @property he defines --background-color with a colour syntax, inherits, and an initial value of white, then applies it to html and main.
• Style queries act as an if statement on a custom property's value. His demo turns a component hot pink when --background-color is black. He was unsure of support at the start, and a reference page said Firefox lacked it. Live documentation then said every element can be a style query container, and he concluded it should work in every browser today. He says it is limited to custom properties for now, with the goal, "I believe", of supporting any property. Giard suggests a use: when users set their own theme colours, the components can keep contrast automatically.

2024-07-29 · open the original ↗

• Responsible AI is a social problem, not only a technical one. Frost, a senior developer at CREMA in Kansas City who sits on the Center for Practical Bioethics' Ethical AI Council, says privacy, bias and value alignment can only be handled by a socio-technical approach that puts "a lot of different people in the room". Giard argues technology is just a tool for good or bad people; Frost agrees, but says AI moves fast enough that its consequences can reach further than earlier tools.
• AI's history shows we may be measuring success wrongly. Frost traces the idea of an artificial person back to the myth of Talos (around 750 BC), 1500s clockwork automata and Vaucanson's 1700s digesting duck, before the term was coined in 1956. She asks whether replicating humans, as the Turing test, chess, Go and sonnet-writing do, is the right goal, and argues AI should augment what people do rather than replace things that are inherently human, like creativity.
• Fear of job loss and AGI crowds out today's problems. She says some job displacement always happens and that forecasts of what will be replaced have historically been wrong. Giard adds his own career advice: watch which skills will be in demand five years out, though he grants AI's pace makes retraining harder.
• Governance should be proactive: consult domain experts before coding. Frost says a developer building a healthcare app needs doctors, health-tech people and legal input as stakeholders, thinking about impact at one, five and ten years. Governance to her means guardrails, measuring fairness and bias, knowing data provenance and building explainability in up front. She expects the rest of 2024 and the next few years to be "a wild ride" for regulation, with state rules from Colorado and California putting borders on technology that has none.
• LLM bias is an error worth understanding, not reality to accept. She cites "mid twenties" of fairness metrics, a number she hedges, and says no single fix exists. Her examples: a story about an engineer comes out male, and one about an engineer named Michelle turns into a story about a woman in a male-dominated field. When Giard says such stereotypes are grounded in the data, she answers that prejudice is a human error and that understanding the errors we feed a model is how it could "do better than us".

2024-07-26 · open the original ↗

• Expert readers do not read books linearly. Matuschak cites ethnographic studies of highly cited professors, who skip from contents to index, preface, chapter openings and figures before choosing what to read; people with reading-strategy skill but no domain expertise can still comprehend hard texts by drawing inferences across distant passages.
• Lectures and podcasts transmit attitudes better than facts. He argues a lecture is a poor way to acquire specifics such as how to calculate torque, but a podcast gives a "portal" into how someone thinks, building tacit scenario templates; a single sentence can change a belief only by tipping a cascade of the listener's own reflection.
• Quantum Country embeds spaced-repetition review in the text itself. The quantum-computing book he built with collaborator Michael Nielsen poses quick questions as you read, then emails you days later to review for a few minutes; he says recalling material when you can only just do so repeatedly extends how long it is retained (the Flappy Bird image is Hanselman's).
• Base knowledge still matters even if AGI arrives. Asked by Hanselman why his son needs logarithms, Matuschak gives two reasons: you often cannot know what to look up without already holding the pieces, and fluency frees working memory ("seven plus or minus two"); a physics student he tutored was struggling because algebra was not yet automatic.
• Many linked representations beat matching a "learning style." He says the evidence does not support learners being fixed visual or symbolic types, but understanding grows with each representation learned and linked; teenagers remember Pokémon better than the periodic table because of richer connections, not choice, and he argues the real gap is having no visceral reason to know.

2024-07-26 · open the original ↗

• Starlink Mini is the dish for mobile use, not fixed sites. The host, who says the video is not sponsored and that he bought the dish and service himself, rates it superior to every earlier Starlink by almost every measure for RVs, camping and off-grid work, but still recommends the full-size residential dish at a home or business for its extra bandwidth.
• The Mini costs more up front than the full-size dish now does. The dish is $599 plus the $20 shipping he paid, while SpaceX cut the standard dish to $499, with promotions as low as $199 in places. Mobile Regional is $150 a month unlimited, in-motion up to 10 mph; Mini Roam is $50 a month with a 50 GB cap and opt-in $1/GB overage; both can be paused and are prorated.
• The in-motion speed limit may not be strictly enforced. The host ran a Gen 3 dish on the Mobile Regional plan through a 12-hour road trip at around 80 mph without trouble. The captions state Mini Roam's in-motion terms ambiguously.
• The router is built in, and unplugging Ethernet breaks the IP67 rating. The dish (12 in by about 10.25 in, just over 3 lb) carries a 3x3 MU-MIMO Wi-Fi 5 access point, which the host argues is enough because internet speeds never reach Wi-Fi 6 rates. Its RJ45 port lets it act as a WAN-only device, but Starlink says IP67 holds only with its own Ethernet cable, unreleased at recording. The data sheet gives 12–48 V input, with a 12 V cable and a USB-C-to-barrel cable "coming soon".
• His backyard tests showed strong speed at half Gen 3's power. Off an EcoFlow Delta, setup took about 2 minutes plus roughly 15 minutes of two firmware updates. The Starlink app measured 138 Mbps down, 16.4 Mbps up and 54 ms; Ookla measured 82 down and about 12 up, against the 150–300 Mbps he typically gets on the full-size dish. Idle draw was 20–30 W (Gen 3: 40–60 W), rising about 10 W, to roughly 40 W, during a speed test.

2024-07-26 · open the original ↗

• Bitnami gives developers ready-made open-source building blocks. Salmeron, a staff engineer on the Bitnami content team (VMware, now Broadcom), describes a free catalog of more than 200 applications and components, including PostgreSQL, MariaDB, Redis, MongoDB and WordPress. It ships as OVAs, cloud images, containers and Helm charts, so a developer can run something like helm install for Postgres instead of configuring it by hand. He says that historically cloud providers paid Bitnami for marketplace listings while users were not charged.
• The paid Tanzu Application Catalog serves platform operators. Salmeron says enterprises need control of licences, software bills of materials, known vulnerabilities and compliance checklists such as STIG, which he calls "a very, very painful process" to do by hand. The open-source catalog uses only Debian, while Tanzu lets a company build the same catalog on Red Hat 9, Ubuntu 22 or its own hardened base image. He points to Log4Shell, when many companies could not tell whether they were affected, as the case for per-component SBOMs.
• Operators curate the catalog and developers pull from a private repository. An operator picks the applications and operating system and sets a target repository. The system pushes container images and Helm charts there in OCI format, and developers use it as they would Docker Hub. He says the charts follow community security checklists, so tools like kubescape and kube-bench report the essential security features as configured, and OVAs are available for teams not on Kubernetes.
• Vulnerability-exchange documents separate scanner noise from real exposure. Salmeron says scanners such as Trivy and Grype often return "terrifying" lists that users cannot interpret. Bitnami bundles analyses saying which reported CVEs actually affect each application, so five critical findings can drop to zero once the document is applied.
• An in-house operating system can get closer to zero CVEs. He says the team cannot do much about Debian, Ubuntu or Red Hat because it does not own them, but has full control over Photon OS, a lean in-house image. He recommends it to vulnerability-sensitive users as part of the commercial Tanzu offering, his employer's product.

2024-07-23 · open the original ↗

• Source generators make code generation part of compilation itself. Bock, a staff software engineer at Rocket Mortgage after almost 25 years as a consultant, says that unlike older tools such as CodeSmith or T4 templates, a generator plugs into the Roslyn compiler pipeline. It runs whenever code changes in Visual Studio or Rider and also under dotnet build at the command line. It can work from the project's existing C# or from "additional files" such as a JSON data file. The feature is C#-only and purely additive: it can add code to a project but never edit existing code.
• The output is ordinary C# you can read and debug. Bock counts this as an advantage over IL weavers such as Fody, which rewrite the binary and need knowledge of intermediate language. Generated code is compiled into your assembly and you can F12 or step into it. By default it is not written into the project folder, which he thinks keeps it out of the Git repo; a project-file setting saves it to a chosen directory. Anyone who clones the repo and builds gets it generated afresh.
• Consuming a generator is easy; authoring and packaging one is not. Using one, such as his own compile-time mocking library Rocks (an alternative to frameworks like Moq), means referencing a NuGet package and adding attributes. Writing one means implementing a single interface, IIncrementalGenerator, with a single method, but it requires understanding syntax trees and semantic models and building C# inside strings. He relies on raw string literals for this, which he thinks arrived in C# 10, and on Microsoft's helper libraries for input/output tests. Packaging has "weird little idiosyncrasies": any dependency the generator itself needs at compile time, and that dependency's own dependencies, need special handling. His lesson is to minimize generator dependencies, and he points to Microsoft's two "C# source generator cookbook" documents.
• A badly built generator slows the compiler, so design for caching. Bock says to avoid the original ISourceGenerator interface, still referenced three years on, because it did not plug into the pipeline well. With IIncrementalGenerator, if the generator builds an immutable model with value equality (records give this out of the box), the compiler skips regeneration when a change yields an identical model. Without that, the generator regenerates the same code on every change. He also says never to make REST or other network calls in a generator: relative to the compiler they are a big performance hit, and they can fail.
• .NET itself now ships generators, the regex one being his showcase. Marking a partial method with what he recalls as a "generated regex" attribute makes the generator write the regular expression's implementation. He says this is typically faster than runtime regex, is AOT-friendly because unused generated code can be trimmed, and gives a hover description of what the generator thinks the pattern does. Its source uses tricks such as goto statements for speed. He also cites community dependency-injection generators that surface IoC registration mistakes at compile time rather than runtime. To learn more he recommends the cookbooks, a Discord channel where Microsoft compiler-team members answer questions, and reading the source of the .NET regex generator, Rocks or others.

2024-07-22 · open the original ↗

• Copilot summarizes a long Outlook email thread in one click. Clicking Summarize on an open thread produced highlight bullet points covering weeks of back-and-forth. In the demo they came out in reverse chronological order, which Giard says is the order the thread itself is in. He pitches it for threads forwarded to you that you would otherwise read end to end for context.
• Draft with Copilot writes a new email from a prompt. On a new message, the Copilot button on the message ribbon has a Draft with Copilot option, available if Copilot is installed and you subscribe to it. You describe the email, including the tone and what not to promise, then click Generate. Giard says the more specific the prompt, the better the draft.
• Drafts can be regenerated, adjusted, then kept or discarded. Regenerate reruns the same prompt and lets you flip between versions. Adjust changes the criteria: longer, shorter, more formal, more direct, more casual, or even a poem. Keep copies the chosen draft into the email without sending it. Discard throws it away.
• The sender, not Copilot, stays responsible for what is sent. Giard's point is that it is "a co-pilot", not an autopilot: you can still edit the draft, you should read it before sending, and you cannot blame Copilot for a bad email.
• Coaching by Copilot critiques a draft under three headings. Also on the ribbon's Copilot button, it gave feedback under tone, reader sentiment and clarity, with suggested rewrites that were mostly more direct or more concise wording. Giard says the thumbs-up and thumbs-down buttons feed back into the engine, which refines its responses if many users rate them down.

2024-07-22 · open the original ↗

• The monitor was supplied free; the review is unpaid. Hanselman says he normally reviews only hardware he bought himself, and made an exception because BenQ pitched the RD280U as a monitor specifically for programmers.
• The 3:2 aspect ratio buys more lines of code. At the same width as his 27-inch 16:9 Dell (3840×2160), the 28-inch BenQ runs 3840×2560, which he calls "4K plus": 400 extra vertical pixels at the same DPI, which he estimates at roughly 20–25% more height.
• Contrast is the most immediate difference he sees. With no calibration, he says blacks are black and dark mode is genuinely dark, while his seven- or eight-year-old Dells look washed out beside it.
• Coding presets tune the panel for dark or light themes. A front-accessible button cycles coding modes for dark and light editor themes; he also shows "Moon Halo" bias lighting on the back, DisplayPort, HDMI and USB-C inputs, a USB hub with KVM, and that it hung on his existing VESA wall arm without changes.
• His verdict is provisional and the price reasonable. He calls $599 a significant upgrade for the money, and says he will use it for a couple of months, judging eye strain, colour temperatures and the bias lighting, before giving an update.

2024-07-19 · open the original ↗

• Good Copilot prompts give context and keep the conversation going. Garcia, managing director of Xebia's US Microsoft services line, says "give me a loop" gets a generic loop. Stating your role, the application's tiers and microservices, the target (Azure), the language and the goal gets an answer fitted to the scenario. He says not to accept the first answer but to follow up ("yes, and also this") until it is right, and that the main way to learn is to keep trying rather than give up after one miss, with GitHub's Copilot enablement videos as a supplement.
• His most useful prompts scaffold, explain or debug code. His examples: ask Copilot to scaffold a cloud data-fetching utility for Azure in a named language, and it builds the whole file structure; ask it to explain a coworker's five-year-old code instead of spending a day reading it. In his own recent code it found an infinite loop he could not locate and pointed to a variable that needed initializing. Giard adds his own use: a starter for a language or library he has never used.
• Copilot's context depends on where it runs and which licence you have. In VS Code it learns from the files you have open. Garcia says Copilot Enterprise on GitHub.com sees the whole repository, so slash commands such as /doc, /explain, /tests and /generate follow the codebase's naming conventions and reuse its existing pieces, such as connection strings. He describes individual, business and enterprise tiers, with enterprise adding controls such as blocking suggestions that could raise copyright problems.
• Copilot won't replace developers; developers who use it will pass them. Garcia says the developer stays the pilot, setting intent and architecture, and that Copilot will not turn a junior developer into a senior one. He compares it to the modern IDE of about 25 years ago: optional, but faster. He predicts that in a year or two developers who have embraced it will get the jobs others do not. If it gets you "maybe 75%" of the way, he says, you spend the rest on the business problem, a point Giard backs with his own story of a manager who said good developers should write the code themselves.
• Copilot is heading into security fixes and browser-proposed pull requests. Citing what GitHub and Microsoft showed at Build "like a month ago", Garcia expects Copilot to work with GitHub Advanced Security to fix flagged issues. He also describes GitHub Copilot Workspace, where you describe a change in the browser and Copilot proposes a pull request that you approve.