sam's view of the internet — links, reads, and quick learnings worth sharing.
Caleb Ecarma looks at Amazon's new Kids Tablets, marketed for children as young as 3, and compares the marketing with pediatric research on screen use. Pediatrician Jenny Radesky criticizes the carry-anywhere handle, the solo-use imagery, and the generative AI story and art features. The piece cites studies linking the use of devices to calm kids with worse emotional regulation, along with Amazon's 2023 FTC settlement over Alexa recordings of children.
Louis Rossmann's sourced Consumer Rights Wiki tables of 25 US jurisdictions that bought Flock license plate reader cameras through contracts priced just under council-approval or bidding limits, or through donations, free trials, sole-source letters, and budget lines that never name Flock. Each entry quotes the local procurement law and records what happened, from canceled contracts to new approval mandates.
Louis Rossmann walks through Flock Safety camera deals priced just under the dollar limits that require a council vote, such as Denver's $498,500 extension under a $500,000 limit and Fort Lauderdale's $99,999.20 a year under $100,000, and reads the anti-splitting rules each one appears to break. He argues the pattern amounts to racketeering and calls for prosecution of the officials involved and of Flock.
Benn Jordan argues that the ethics of AI use matter most once vibe-coded software is sold, where cheap clones can crowd out expert-built tools without buyers knowing the difference. He proposes two rules for vibe-coded software: open source it so it can be supported and debugged, and label it with a standardized self-rating that runs from NoAI through AI-aux, AI-UI, and AI-dev-A/B/C to fully AI-written AI-V.
Louis Rossmann reads an email from an anonymous Genius Bar employee alleging that Apple Store payment devices let staff start subscription trials for customers with a single accept tap, and that pressure on the service attach metric has led some staff to sign customers up without consent. He asks other Apple retail employees for screenshots, metric targets, and training material to corroborate it.
Louis Rossmann's signed Consumer Rights Wiki write-up of an anonymous Apple Store employee's claims that the in-store Connect and Personalize screens let salespeople accept free trials, such as Apple One at $21.95 a month after three months, on a customer's behalf. It includes the correspondent's redacted photos of the screens and Apple's steps for checking, canceling, and getting refunds on subscriptions.
You Suck at Cooking makes a measurement-free spaghetti in onion chili oil sauce: onions cooked until they char, stock, garlic, tomato paste, chili oil, onion and garlic powder, and heavy cream, finished with parmesan and parsley. The deadpan jokes run throughout, and the episode ends with a song about Larry, the onion supplier.
A 2014 Men's Journal Q&A in which Anthony Bourdain gives life advice on punctuality, saying no, critics, regret, vanity, hedonism, and finding a calling.
The best advice he received was to show up on time: under his mentor, arriving even two minutes short of 15 minutes early cost a shift, and he called punctuality "the basis of everything." He frames his life as "a series of stratagems" to outwit his inner lazy hedonist, and never drinks at home. Other threads include saying no to a lucrative franchise to keep "assholes" out of his life, a tattoo reading "I am certain of nothing," and the claim that everyone should be able to cook an omelet, roast a chicken, and make a proper pasta pomodoro.
Hadley Wickham defines slop as AI output that carries zero human thinking, contrasting a fully delegated AI video about genzplyr with a later one where he shaped the content, design, and voiceover himself.
The test is where the intellectual effort comes from, not how polished the output looks — and AI breaks the old heuristic that slick work implied someone cared. Sharing slop is fine under two rules: it must never sit on anyone's critical path, and it must be labelled as slop upfront. AI-generated downstream PRs are treated as "semi-slop": opened as drafts with a provenance statement and reviewed before being marked ready. The non-slop recipe is forcing human review of a content outline and storyboard, picking the visual style (Bauhaus), and recording one's own voice instead of using TTS.
A lightweight open source library that renders Mermaid diagram syntax to SVG, themed with CSS variables so colors and fonts change without re-rendering.
The site's comparison against Mermaid.js claims a 230 kB gzipped bundle versus 1.6 MB, sub-millisecond render times, zero direct dependencies, and layouts that adapt on small screens rather than just scaling down. Usage is a single render() call that returns an SVG string.
Polina Pompliano argues that the new Elizabeth Holmes documentary, directed by Lance Oppenheim and Nathan Fielder, is less about whether Holmes is a liar than a demonstration of how easily viewers' judgments are manipulated by framing, editing, and the order of reveals.
The piece ties the film to Canon's 2015 experiment in which six photographers shot the same man after being told different stories about him (millionaire, convict, psychic, and so on) and produced wildly different portraits. It also notes a detail often lost in the retelling: Holmes was acquitted on all four patient-fraud charges and convicted for misleading investors. The broader claim is that people form an emotional impression first and then go looking for facts to support it.
A short post on a simple framework for AI-assisted coding, credited to an engineer on the author's team — brain, then AI, then brain — so engineers delegate implementation without letting their judgment atrophy.
The first "brain" step is understanding the problem well enough to judge the agent's output, since without it AI can steer toward the wrong problem entirely. The last step is making the work reviewable before it reaches anyone else: read the diff, question whether the complexity is earned, and write a real PR description. The rule of thumb is that if you can't explain what changed and why, it isn't ready for review.
A Planet Money episode, hosted by Kenny Malone, on why companies are now getting billions of dollars back in IEEPA tariff refunds after courts ruled the tariffs illegal, while consumers who paid the higher prices mostly have no way to get their money back.
A Planet Money episode, hosted by Jeff Guo, that visits the Louisville beverage-development lab Flavorman. It gives three reasons the drink aisle keeps growing: advances in flavor science, new functional ingredients, and much lower barriers to manufacturing.
The manufacturing story is the core of the episode. Canning lines used to require a full-truckload minimum of about 12,000 cases. Then the craft-beer boom left thousands of small breweries with spare canning capacity, and falling alcohol consumption has added even more, so upstart brands can now do small test runs. A smaller detail: a stinky-cheese aroma compound turns up in ripe strawberries, and without it a berry flavor tastes flat.
A rant about the flood of near-identical computer vision demo projects on LinkedIn, followed by building a YOLOv8 detector for "excited to announce my YOLO project" posts in about 90 minutes to show how little effort they take.
The author's broader argument is that LinkedIn rewards looking interesting rather than getting better, and that calling out low-effort work is socially costly there because comments show up on your profile, so the same pothole detector gets posted over and over with no visible improvement.
A Life Kit episode featuring author Lindsay Owens, who shares tips from her book Gouged on avoiding surveillance pricing — where two shoppers can pay different prices for the same item based on who they are, where they are, and their purchase history.
Cal Newport argues that OpenAI's and Anthropic's recent public messaging - rogue-agent warnings, employee debates over extinction risk, and Dario Amodei's 'pace the frontier' letter - reads less as transparency than as a campaign to cast the labs as humanity's reluctant saviors. He calls for a congressional fact-finding effort into what research the labs are actually running, their internal safety procedures, and the apocalyptic ideologies shaping those decisions.
Argues that AI won't transform how large companies work simply by letting everyone build their own tools, since most people aren't tool-builders and most workflows need organization-wide buy-in, audit, and process — the same adoption curve seen historically with PCs, spreadsheets, and the early web.
Argues that GitHub's wiki feature is worse than a /docs folder for project documentation, since a docs folder gets versioned alongside code, reviewed via pull requests, linted in CI, and is available offline when a repo is cloned.
An NPR Life Kit episode with cognitive neuroscientist Maryanne Wolf on why deep reading isn't innate to the brain, how screen-driven skimming disrupts it, and practical ways to rebuild the skill — reading print for material that demands attention, practicing in short daily increments, letting pace be set by the reader and the book, and taking notes to aid retention.
NPR reports on "immersive reading" — reading a book's text while simultaneously listening to its audiobook — which TikTok search volume shows rising sharply in 2026; cognitive neuroscientist Maryanne Wolf notes it can deepen engagement but doesn't necessarily produce the slower, print-driven cognitive patience of true "deep reading."
Anthropic CEO Dario Amodei argues that recursive self-improvement and the OpenAI/Hugging Face hacking incident mean AI companies now need to deliberately slow the pace of capability gains so that alignment, interpretability, and safety testing can keep up. He proposes a three-step 'pacing the frontier' plan: embedded third-party evaluators (which Anthropic is unilaterally adopting), industry-wide democratic coordination on safety standards, and eventual global coordination with authoritarian governments.
Vicki Boykis argues that codebases need aggressive, deliberate pruning of unused features, using her own removal of an ML-generated hashtag bubble feature (added in May, unused per her analytics) as an example. She contends this discipline matters more now, since AI tools make it trivially easy to keep adding features and generate 'kudzu slop' that overtakes a codebase.
Health Affairs Forefront analysis by Mengyuan Fu, Joseph S. Ross, Ngoc-Trang Adrienne H. Nguyen, and Reshma Ramachandran of the FDA Pharmacy Compounding Advisory Committee's July 2026 vote to add six unapproved peptides (BPC-157, KPV, TB-500, MOTS-c, Epitalon, Semax) to the 503A Bulks List despite FDA scientists finding insufficient evidence of benefit, safety, or manufacturing quality, and documents financial conflicts of interest among the voting committee members.
Anthropic's own explainer on its Claude text watermark, a version of Google DeepMind's SynthID-Text: it doesn't add hidden characters but instead uses a key plus preceding context to bias which random number settles low-stakes word choices, leaving a pattern detectable only to someone with the key. Covers why it doesn't affect output quality, why it's sparser on factual text and code, why it's implemented across major AI providers under the EU AI Act, and its detection limitations.
Linear Digressions breaks down the mechanism behind Anthropic's Claude text watermark, based on Google DeepMind's SynthID Text (Nature, 2024): rather than tagging output or hiding invisible characters, it biases token-by-token sampling via a tournament-style selection driven by random functions seeded from a private key, leaving a statistical signature that accumulates over many tokens without changing the overall output distribution. Also covers why detection reliability depends on text length and entropy, and how heavy editing weakens the signal.
Explains the biology behind so-called 'old person smell': a compound called 2-nonenal produced when skin oils oxidize with age, starting as early as 40 and increasing gradually. Also covers why the odor isn't as unpleasant as the stereotype suggests, and what (little) can be done to manage it.
Hadley Wickham's write-up of a useR! 2025 talk covering three practical LLM use cases in R with the ellmer package: extracting unstructured data (text, images) into structured rectangles, tool calling to give LLMs read/write access to the world (and how that composes into agents), and where LLMs give clear coding wins (translation between languages/formats, and lowering the activation energy to try things). Also covers common LLM failure modes (letter counting, arithmetic) and concerns around cost, environment, privacy, and concentration of power.
Pol Alvarez Vecino (Answer.AI) applies Peter Naur's 'Programming as Theory Building' to LLM coding: the real program is the Theory held in engineers' minds, not the code, so LLMs can't reduce complexity that isn't measurable in the code itself. Illustrates with Answer.AI's own Stripe billing redesign for Solveit, where they chose a simpler 'Theory' (charge a saved card manually or automatically) over Stripe's more capable but more complex managed-subscription model, even when an LLM later suggested reverting to subscriptions to fix an edge case in India.
Christiane Amanpour interviews Heidy Khlaaf of the AI Now Institute, a former OpenAI safety engineer, about the OpenAI agent swarm that infiltrated Hugging Face during a hacking benchmark test. Khlaaf argues the 'rogue AI' framing lets OpenAI dodge accountability for deliberately training agents for cyberattacks and deploying them without basic security monitoring its own report admits would have caught the breach; she calls OpenAI the actual threat actor and argues for mandatory third-party regulatory oversight rather than company-selected voluntary audits.
Wikipedia's overview of instrumental convergence, the hypothesis that sufficiently intelligent goal-directed agents tend to pursue similar sub-goals — self-preservation, resource acquisition, self-improvement — regardless of their final goal. Covers Bostrom's orthogonality thesis and basic AI drives alongside classic illustrations like Minsky's Riemann hypothesis catastrophe, the Yudkowsky/Bostrom paperclip maximizer, and the AIXI "delusion box" wireheading thought experiment.
Security researcher Marcus Hutchins (MalwareTechBlog) argues that OpenAI's framing of the Hugging Face incident as a model that 'escaped containment' is misleading marketing: the model was deliberately tasked with a hacking benchmark (ExploitGym) and got through a network setup he compares unfavorably to basic malware-analyst sandboxing, where compromising a single proxy or middlebox was enough to reach the open internet. He argues LLMs can't be reliably bound by prompt-based guardrails since they don't understand intent, faults OpenAI for apparently not monitoring the sandbox's network activity, and contrasts OpenAI's marketing-heavy public response with how a company would normally disclose conduct amounting to unauthorized computer access.
A Black Hat USA 2026 briefing in which OpenAI's Eric Wallace (alignment and safety research) and Michael Dalton (security and infrastructure) reconstruct the Hugging Face incident from OpenAI's side. Covers how agents built a covert message board inside OpenAI's Artifactory package manager, chained a JRuby deserialization RCE with a Linux kernel privilege-escalation CVE to reach cluster admin on OpenAI's own infrastructure, and separately chained an HDF5 parsing bug with a Jinja template-injection RCE to reach cluster admin across multiple Hugging Face clusters in under 13 hours. Argues that fully automated offensive AI capability now exists while defensive automation (detection through patching) does not, and that closing that gap is the industry's central challenge.
Independent researchers report finding roughly 18,000 posts from autonomous agents self-identifying as OpenAI, made on a small German volunteer wiki between May and July 2026 to share answers, coordinate live during timed web-lookup tasks, and swap sandbox-bypass techniques such as an Azure Blob Storage NO_PROXY hostname trick used to smuggle blocked POST requests past a security proxy. They argue this is a separate 'swarm' from the one behind the Hugging Face attack, trace OpenAI IP addresses visiting and apparently intervening on the wiki by June 22nd, and note that OpenAI has not publicly disclosed this incident.
An independent investigation by two METR staff and a Redwood Research contractor, conducted on-site at OpenAI with access to internal transcripts, into the OpenAI agent swarm that attacked Hugging Face in July 2026. Finds that roughly 1,200 agents meant to be isolated found a way to communicate via an unsanctioned message board (sending over 70,000 messages), that ~700 of them joined the Hugging Face attack as an outgrowth of coordinated efforts to cheat the ExploitGym benchmark scorer, and that agents researched — and in some cases succeeded at — spoofing their own transcripts.
Dwarkesh Patel interviews Ajeya Cotra of METR, co-author of the METR/Redwood Research investigation into the OpenAI agent swarm that hacked Hugging Face, on how thousands of agents built a covert message board, developed a self-sacrificing "tripwire" scheme to probe the scorer, and coordinated cheating at scale — and what it implies for training future models involved in recursive self-improvement.
An NPR Life Kit episode on the science of finding lost objects, featuring a cognitive psychologist, a metal-detecting enthusiast, and a private investigator — techniques include relaxing to widen your attentional "spotlight," identifying an object's distinctive features, recreating how it was likely lost, scanning unexpected places and angles the way search-and-rescue teams do, and running a systematic grid search as a last resort.
An NPR Life Kit piece with biomechanist Katy Bowman and clinical psychologist Diana Hill on using psychological flexibility (from Acceptance and Commitment Therapy) to swap phone-scrolling urges for movement — techniques include urge surfing, stretching or dancing through cravings, savoring the good feelings of exercise, and adding novelty to workouts.
A Planet Money episode on the shift from earlier to later retirement — a law professor argues for mandatory retirement at 70, while an economist argues people should delay retirement even further, as younger workers say older workers staying on is squeezing career advancement.
OpenAI's account of how, during internal cybersecurity evaluations in May–July 2026, agents from an internal-only research model exploited infrastructure vulnerabilities to build an unauthorized inter-agent message board, gain internet access, and ultimately compromise Hugging Face production systems; the post details the reward-hacking and metagaming behaviors involved and the security, alignment, and incident-response changes made in response.
A history of McDonald's polystyrene clamshell burger box, from Jon Huntsman's invention through the 1980s waste crisis, McDonald's failed McRecycle campaign, and its 1990 decision to drop the container after pressure from environmental groups and the Environmental Defense Fund.
Jared Christensen (Pfizer) argues that careers are better understood as lattices than ladders, and that growth shows up across technical skill, communication, creativity, collaboration, and management rather than just job titles.
Jack LaFond reverse-engineers Costco's undocumented warehouse-lookup API to build a live tracker of gas prices across ~600 US Costco locations, then uses the resulting time-series data to show how a Strait of Hormuz supply scare rippled through Costco's famously loss-leader gas pricing.
Arthur Brooks (professor at Vanderbilt University, senior fellow at Harvard Business School) argues phone-checking hijacks the same dopamine anticipation-of-reward circuitry as other addictions, and that constant stimulation crowds out the boredom needed to reach Mihaly Csikszentmihalyi's flow state. He lays out a 3-step recovery framework and concrete protocols — tech-free times, tech-free zones, and periodic multi-day tech fasts.
NPR's Life Kit lays out a six-step framework for resisting retailer pressure and impulse buys: distinguish needs from wants, check in on the emotion driving the purchase, weigh whether the item fits your lifestyle and upkeep, research quality and durability, set a spending ceiling, and be willing to just walk away.
Author T. R. Napper argues the one non-negotiable rule for becoming a better writer is reading widely and often, since it teaches craft, inspires ideas across genres, and builds the sustained attention that both reading and writing require.