Love the Machine (Learning)

Since 2022, hundreds of billions of dollars have been spent on making sure the public associates AI with LLMs and not with the real workhorse—Machine Learning. It’s the kind of intelligence that’s been reading radiology scans and flagging irregular heartbeats for well over a decade. It’s narrow, task-specific, lacks the ability to hallucinate, and is reasonably well-validated.

The problem is generative AI: large language models that predict the next plausible word in a sequence and call it knowledge. The same tools that have convinced teenagers to kill themselves, can’t consistently count the number of Rs in “Strawberry”, and are designed for sycophancy and maximum engagement, are now being handed the keys to clinical documentation, patient triage, diagnostic assistance, and drug discovery.

Eighty-one percent of physicians now use AI professionally, according to the American Medical Association’s 2026 survey — more than double the rate from 2023. Meanwhile, consumer trust in AI for healthcare has plummeted, from 52% to 44% over the same window. Generative AI’s obvious failings threaten the reputation of even the most basic of “dumb” machine learning tasks, because how is the public meant to discern between generative AI and ML when their doctor tells them they used “AI” to assist with a diagnosis? Hell, even the doctor might not be able to tell the difference. They get pitched on dozens of tools at any given time, and most don’t understand their inner workings. After all, why should they? They’re a doctor, Jim, not a software engineer.

Excuse Me, Your Privates Are Showing

HIPAA was written in 1996. It covers providers, insurers, and clearinghouses. It does not cover your Oura ring, your Fitbit, your ChatGPT conversation about that weird mole, or the genetic data 23andMe auctioned off when it went bankrupt. A booming market of data brokers now pays hospitals for “de-identified” patient records to train AI models — a market valued at nearly $9 billion — and researchers have already shown that LLMs can re-identify those supposedly anonymous records. One NYU study pulled a specific pregnant patient’s identity from a clinical note that mentioned nothing more identifying than horseback riding.

Google, meanwhile, is assembling the most comprehensive health-data pipeline on Earth, though you probably already assumed that. Android owns roughly 70% of the global smartphone market, and Google also owns Fitbit. They made a promise in 2019 to keep health data out of advertising, which sounds reassuring until you read the actual privacy policy, which permits using data to “improve services” — a phrase capacious enough to drive Sam Altman’s ego through.

Large Language Modus Operandi

The opt-in model (those papers you glanced over at your last doctor’s appointment before signing anyways) is already eroding. AI scribes record your doctor’s visit and generate notes — and while you can technically decline, good luck navigating that conversation mid-appointment when your ass is finally seated after a three hour wait. One patient in Australia had a $1,300 appointment cancelled for refusing. You see where this is going: today’s opt-in becomes tomorrow’s default, becomes next year’s invisible infrastructure, just like browser cookies, just like location tracking, just like every other privacy trade-off we’ve clucked our tongues at while striding full bore into.

With all of that out of the way, this tunnel may be shorter than it seems. World models, an alternative architecture (or the next evolution after LLMs) learn how systems evolve over time rather than predicting the next word. They don’t hallucinate in the same way because they aren’t generating text — they’re modeling states.

Early research is applying them to tumor progression, disease forecasting, and surgical robotics. They’re immature, unstable, and at least a year or more away from any real product. But they represent the only credible path toward AI that a clinician could somewhat-reliably trust with a diagnosis. Where chatGPT takes a stab at statistical correlation, a world model is purpose-built for prediction, rather than prediction as (and I’m being slightly reductive here) an afterthought.

Half-Baked

The health industry is adopting AI faster than its trust infrastructure, its privacy law, and its cybersecurity posture can support. That’s straightforwardly a governance problem, and governance moves at the speed of committee hearings and election cycles while technology moves at the speed of venture capital.

So what does the health market look like when the keys get handed to generative AI? Probably a lot like the internet looked when we handed the keys to ad tech: useful, ubiquitous, and corrosive to the things we said we cared about most. The difference is that this time the thing being corroded isn’t your attention, it’s your medical records, your literal genetic code, and your trust in the person across the exam table telling you those cells probably aren’t cancerous.

As always, zack.wall@icloud.com for feedback and suggestions.

Vomit On My Sweater Already, Will’s Spaghetti

Remember that cursed video of Will Smith eating spaghetti? I’m going to hold your hand when I tell you that it’s been more than 3 years since it was generated. Time flies when you’re waiting for our AI overlords to come online! The video was funny because it was terrible—surreal hands, noodles fusing with skin, a face melting into marinara. Nobody was fooled, and that was the entire point. AI video was a party trick—something you showed your friends so you could all laugh at the absurdity.

That innocent era is over.

In less than three years, AI-generated imagery has completed a full transformation in purpose. It is no longer a creative tool, it is overwhelmingly a deception tool. The product has shifted from “look at this obviously AI-generated art” to “look at this real thing that happened.” The people making that shift have clear agendas: engagement farming, political manipulation, romance scams, and outright commercial fraud.

Oops, No Guardrails!

When OpenAI launched Sora 2 back in September 2025, it became the most downloaded free app in the App Store within a week. Every output carried a visible watermark and invisible C2PA metadata. Both protections collapsed almost immediately. Watermark-removal tutorials flooded YouTube and TikTok within days. The Pro API tier shipped outputs with no watermark at all (almost like they knew why people were paying for the Pro tier). By March 2026, OpenAI shut the entire Sora app down, citing a number of reasons (the most likely was costs and a lack of financial return). The tool’s primary legacy wasn’t creative expression. It was industrialized lying.

NBC News documented Etsy storefronts selling crochet patterns marketed with AI-generated images of plushies that real yarn physically cannot produce. Even recipes haven’t been spared, with CNN reporters being duped into baking a recipe from Pinterest that turned out to be entirely AI-generated and didn’t work. Pinterest’s CEO conceded on an earnings call that no platform can catch 100% of AI-generated content. The marketplace has become a minefield where the product photos are fiction and the patterns are impossible.

On a more dangerous topic, Sumsub’s 2025-2026 Identity Fraud Report found that deepfake fraud now accounts for 11% of all global fraud, with dating apps tied as the single most-targeted sector. A French interior designer handed over €830,000 to scammers running an AI-generated Brad Pitt impersonation. The FBI’s 2025 IC3 report logged roughly $20.9 billion in total losses, with romance fraud among the costliest categories and over 22,000 AI-related complaints. These aren’t edge cases. This is one of the primary commercial applications of the technology.

Creatively Bankrupt

There is a reason these tools are being used for deception rather than creation: the creative industries have made it socially radioactive to admit you used AI. SAG-AFTRA struck for eleven months over AI in video games. Studios have had titles canceled over AI asset backlash. Wizards of the Coast was caught using AI art for Magic: The Gathering despite publicly pledging not to. The message from Hollywood laborers, gaming, and the art world is clear—use AI openly at your own risk.

That stigma is obviously earned. AI art should remain a personal tool—something you use for fun to share with friends and family. It has no place in commercial creative work where a thing with no soul, no experiences, no taste, no worldview, no self-awareness is displacing real human artists. The unfortunate consequence of this is that the path of least resistance becomes deception. Passing AI output off as a real photo, a real person or a real product carries no professional penalty if it’s never caught.

Every Man For Themselves

The regulatory response is basically a finger wag and a stern look. The EU AI Act’s watermarking mandate doesn’t take full effect until August 2026. The United States has no comprehensive federal labeling law. Platforms rely on an honor system where uploaders self-disclose AI content—a system that scammers defeat by stripping metadata and cropping watermarks. TikTok can host every AI-generated video in existence without ever being required to prove whether it’s real. C2PA metadata, the industry’s best technical solution, is destroyed every time an image is laundered through WhatsApp, iMessage, or Facebook’s upload pipeline.

The technology will only get better. The regulations will continue to not exist. And every day, the gap between what AI can fabricate and what platforms can detect grows wider. We moved from Will Smith eating spaghetti to industrialized fraud in under three years. The next three will be even worse.

As always, zack.wall@icloud.com for feedback and suggestions.

A Strait Mess

The Strait of Hormuz has been closed since March 2, 2026. Brent crude hit $126 a barrel in April. QatarEnergy declared force majeure on all LNG shipments within days of the closure. Analysts and much of the public at large believe this to be the largest energy crisis in recent history, perhaps in all of history. In fact, most of the consequences of this catastrophe won’t be felt until 2027, when increased prices on commodities such as helium and fertilizer start to affect the food supply chain.

Meanwhile, the world added 692 gigawatts of renewable energy capacity in 2025 — the largest single-year buildout ever recorded. China alone accounted for 440 GW of that. India crossed 50% non-fossil installed capacity five years ahead of its Paris commitment. The US, by comparison, added a modest 55 GW of solar, wind, and storage, with projections of 86 GW more in 2026.

So the question feels obvious: will 2027 be the year of renewables? Contracts are being signed now. Capacity is being built now. The oil crisis is happening now. Is this finally happening?

Ehh…maybe?

Hyperscalers Are Hyperpolluting

If you only looked at the numbers, with global renewable capacity hitting 5,149 GW at end of 2025, solar PV alone meeting over 25% of global energy demand growth for the first time ever, battery storage deployments up 40% year-over-year, you’d think the transition was a done deal. And on the infrastructure side, it would be silly to argue against the trajectory. China is installing roughly 100 solar panels per second and a new wind turbine every 10 minutes! You don’t even need to be a propagandist to recognize how far the US is falling behind based on those numbers.

China hit their official 2030 target of 1,200 GW wind-plus-solar six years early. Europe’s REPowerEU plan targets 42.5% renewables in final energy by 2030. The US, even with legislative headwinds, is still building almost exclusively clean capacity.

But the companies burning the most electricity on the planet right now — the AI hyperscalers — are not following that curve. Meta signed a deal with Entergy Louisiana for seven new natural gas plants totaling 5.2 GW to power its “Hyperion” data center.

Natural gas plants in the year of our lord 2026!

Microsoft is in exclusive talks with Chevron for a 2.5 GW gas plant in West Texas. OpenAI and Oracle’s Stargate complex in Abilene is running on 4.5 GW of natural gas turbines.

Corporate clean energy purchasing actually fell 10% in 2025 — the first decline in nearly a decade — even as Amazon, Meta, Google, and Microsoft accounted for nearly half of all such deals globally. So much for the 2010s kumbaya save-the-planet train. The message from tech giants is clear: we’ll buy the green credits but we’ll burn the gas, thanks!

The Green (Cash) New Deal

And about those green credits. Every major hyperscaler made bold climate commitments at the start of this decade. Microsoft pledged to be carbon negative by 2030. Google set an ambition for net-zero emissions across all operations by 2030. Amazon targeted net-zero by 2040. Meta committed to net-zero for its value chain by 2030.

That’s how it started, and here’s how it’s going.

Microsoft’s overall emissions have risen over 29% since 2020, driven almost entirely by data center construction. Google’s emissions jumped 48% over five years and the company now describes its climate goals as “moonshots” — a word that does an insane amount of heavy lifting when their stated deadline is four years away. The 2025 Corporate Climate Responsibility Monitor, published by NewClimate Institute and Carbon Market Watch, concluded that these five companies’ emissions targets have “lost their meaning and relevance.”

Microsoft, to its credit, has been the single largest buyer of carbon removal credits on the planet — responsible for an estimated 80-90% of the entire market. But even that program reportedly paused in early 2026.

The hyperscalers haven’t exactly abandoned their climate language, they’ve just redefined what words mean.

Okay, And?

Let’s talk about the people. A Gallup poll from March 2026 found that 71% of Americans oppose having an AI data center in their area. Not a controversial take, but what’s interesting is that’s more opposition than nuclear power plants get. When you dig into why people are opposed, it’s not really about carbon, it’s more about how some of these data centers make for poor neighbors thanks to their water, air, and noise pollution, alongside rising electric bill costs and a general anti-AI sentiment.

Switching a data center from natural gas to a solar farm doesn’t fix the noise, it doesn’t reduce the water draw, and it doesn’t stop the grid upgrade costs from being passed to residential ratepayers. On renewables more broadly, public enthusiasm in the US is actually softening. Gallup’s 2026 environment survey found support for greater emphasis on solar dropped seven points and wind dropped eleven points since 2021. Nuclear, meanwhile, is at an all-time high.

Quietly, the electrification of transport keeps accelerating regardless of political headwinds (read: the Trump administration). Global EV sales hit 20.7 million in 2025 — nearly 23% of all cars sold. China is at 55% EV share for new vehicles. Europe grew over 30% to a 28% share. And then there’s Canada, which in January 2026 replaced its 100% tariff on Chinese EVs with a 49,000-unit annual quota. The US called it a disaster (they really hate competition). Canadians see it as progress.

So…2027?

The IEA expected renewables to overtake coal as the world’s top electricity source by 2026 at the latest. On the capacity charts, the answer to the question is almost certainly yes — 2027 will be another record-breaking year for renewable energy buildout, driven primarily by China and increasingly by India.

But the question implies something more than capacity additions. It implies a shift in how people think, how companies invest, and how governments legislate. And on those fronts, the picture is truly a hot mess. The Iran war crisis made energy independence visceral in a way that climate rhetoric never could, but the US is bafflingly accelerating the phase-out of IRA clean energy credits, Europe is still buying Russian LNG, and the companies with the largest electricity appetites on Earth are building gas plants as fast as they’re signing solar contracts.

The infrastructure is moving in one direction, American politics is moving in the other. Whether renewables continue their trajectory in 2027 depends on whether the people demanding change can stay louder than the companies writing the checks.

As always, zack.wall@icloud.com for feedback and suggestions.

Shots Fired

At Google I/O 2026, Sundar and co. unveiled what they call the biggest change to Google’s search box in 25 years. Instead of returning a list of 10 blue links to websites that might answer your question, Google now drops you into their AI-generated summary whether you wanted one or not. If that isn’t enough, Google will soon dispatch “information agents” to monitor the web on your behalf and send you personalized updates so you never have to visit a yucky website ever again. Leave the thinking to the machines!

TechCrunch’s headline said it plainly: “Google Search as you know it is over.”

Nilay Patel (EIC of The Verge) coined the term “Google Zero” back in 2024 to describe the moment when traffic from Google to the open web approaches zero. At the time, it felt provocative, but in 2026 it’s clearly Google’s roadmap. Pew Research studied nearly 69,000 real searches and found that when an AI Overview appears, users click through to a website just 8% of the time compared to 15% without one. Only 1% of users click on a cited source inside the overview itself. For publishers, the math is existential. These aren’t blog posts complaining about algorithm changes. These are the people who run the web’s information infrastructure telling us that the internet’s wild west era is coming to a close.

The Ugly Cascade

Here’s where people miss the full picture. This isn’t just a media problem, although certainly they’ve spent years sanewashing this AI gold rush and perhaps deserve a comeuppance. This goes beyond news sites and independent journalists and points towards a complete ecosystem collapse. Think of it like a local grocery store shutting down. The diner next door loses its lunch crowd. The dry cleaner loses the foot traffic from a busy block. The farmers who supplied the grocery store lose a revenue stream and may go under themselves. The town doesn’t just lose a store — it loses the economic web that the store anchored.

The internet works exactly the same way. A recipe blog loses 70% of its traffic and can no longer afford its freelance food writer. That writer stops subscribing to the photography newsletter that taught her food styling. The photographer drops her hosting plan. The hosting company lays off an engineer. And on it goes.

We already have casualties. Travel blogs, recipe sites, household product reviews and more are being wiped out thanks to AI summaries. Digital Trends collapsed from 8.5 million monthly clicks to 265,000 — a 97% wipeout. Business Insider’s organic search traffic fell 55%, prompting layoffs of a fifth of its staff.

It doesn’t stop at content. Local businesses are vanishing from AI-generated results entirely. One study found that ChatGPT recommends just 1.2% of all local business locations. If you’re a plumber or a bakery or a new restaurant, the way people used to find you — through a Google search that led to a review site or a local blog — is evaporating. What replaces it? A chatbot summary sourced from data that may or may not be current, with no link for you to click and bless them with your money or time.

The Dichotomy Of It All

Here’s the part that should worry Sundar and team, assuming they’re thinking past the next earnings call. Large language models are trained on human-generated content from the web. When AI models are trained on the output of earlier AI models — recursively, generation after generation — they suffer irreversible quality degradation. Researchers call it “model collapse.” Outputs drift toward bland averages. The tails of the distribution (read: the human parts) disappear.

Google needs fresh, high-quality human-written content to keep Gemini useful. But Google’s own products are systematically destroying the economic incentive to produce that content. Kill the publishers, lose the training data. Lose the training data, degrade the model. Degrade the model, lose the product. The snake eats its tail until nothing is left. The irony of human greed destroying artificial intelligence.

Google’s CEO, when pressed about specific publishers that have been devastated, compared them to individual restaurants losing customers — as if the collapse of an entire industry’s distribution model is the same as a slow Tuesday night. Demis Hassabis, the one person at Google with the scientific credibility to sound the alarm, has spent the last year talking about AlphaFold and drug discovery and…(sigh) goddamned AGI. From an outsider’s perspective, it really seems like Google is rushing to the endgame without a fucking care for what happens in the middle of now and building digital Jesus.

One Idiot’s Case For Optimism

Perhaps it’s copium rotting my brain, but I would be lying if I said I wasn’t a bit interested to see how the new internet evolves with this paradigm shift. The printing press didn’t kill communication. Radio didn’t kill newspapers. The internet itself didn’t kill television. Hopefully in 20 years we can look back and say “AI didn’t kill the open web”. I don’t think it will die, but it will change, dramatically and painfully, and the version that emerges will be smaller, more consolidated, and more dependent on direct audience relationships than the one we grew up with. Newsletters are booming, communities are migrating to group chats and Discord servers (not optimal, but noteworthy). Independent creators are building direct, subscription-supported relationships with their audiences that cut out the corporate juggernauts of yore.

If the web is going to survive this onslaught from Google and AI, it won’t be because they suddenly wake up and remember to be good stewards of the web. It’ll be because enough people decided that a search engine shouldn’t get to shape the future of the internet that WE built. Find those weird corners of the web and keep them weird. Build communities that Google can’t gatekeep in their sanitized bubble. Support that open source project you love. Put your friends and loved ones onto alternatives to Google (Kagi search, Proton Mail, Firefox browser, etc). Let not an apathetic spirit be the reason our beloved internet dies.

As always, zack.wall@icloud.com for feedback and suggestions.

We Were So Naive

Depending on your age and the corner of the world you grew up in, it may be difficult to remember life before the internet. I was born in 1991 and for practically all of my formative life, I had access to some form of the world wide web through NetAccess and AOL, then followed by Road Runner and Spectrum when we could finally afford broadband. Some of my earliest memories are of being kicked off the internet during chat sessions with friends on AIM so my parents could make a phone call.

When the internet hype engine took off, it truly felt like the next leap of human evolution. What happens when I can chat with someone across the world in a near-instant? What does society look like with all of human knowledge an Ask Jeeves search away? Will Walmart be destroyed by e-commerce? What the hell is online dating?

Every day in the early 2000s, a new website popped up, a new product was announced, a new internet gadget was unveiled and the hype seemed infinite. Yeah, yeah the Dot-com crash happened. Who cared? Myspace kicked ass, Facebook was up-and-coming, Netflix was going digital, 2-day shipping was magical, and the world felt right.

When we were of the age and knowledge to start shaping Web 2.0, we needed no permission. Geocities, Tumblr, Blogspot, Club Penguin, Newgrounds, Second Life, and thousands more places were created by and for millennials to congregate, discuss, troll, dream, hope, and build.

The Wide Web Narrows

Contrast that to today’s internet, where you can count on two hands the number of websites and apps you open or visit on a weekly basis. Nilay Patel calls The Verge “the last website on earth”, and as much as I love reading The Verge, I think there can be no stronger indictment of the modern web than that.

At some point you logged out of Club Penguin or Facebook or Newgrounds for the last time. “That’s just called growing up.” No. I reject that. We were happy once, and then enshittification happened. When people discovered that they could make money on the internet, the vibes changed overnight.

Consider that Roblox and Fortnite are no longer innocent corners of the internet for kids to experience safe, wholesome fun. They are extraction zones designed to get children addicted to spending and gambling real money on fake digital goods—that’s it.

Just like the disappearance of kid-friendly public spaces, kid-friendly internet spaces are also experiencing an extinction event, and so-called Age Verification laws will be the final nail in the coffin for them. There is a dark irony in the billionaires and Epstein-associates, who are funding and writing these laws, telling us that mass surveillance is required to protect children from predators.

If we sand off every sharp corner that the internet is made of, what exists but an emasculated promise, murdered by surveillance capitalism, manufactured safety, and short-term profit gains?

The concept of the borrowed web has disastrous long-term implications. What has Gen Z built that it truly owns, and how much of what they interact with is actually a rat maze masquerading as a funhouse? The teens are getting more and more creative with how they inhabit the spaces they inherited, and they are genuinely shaping internet culture in fascinating ways, but those aren’t spaces that they own or created.

A long time ago, a millennial college student wasn’t satisfied with MySpace, and so he built Facebook. If a Gen Z programmer isn’t happy with TikTok, who’s going to fund their short-form video startup? Long has consolidation been the enemy of invention. Many smaller ideas merge to become a few dominant ideas, and then the people in charge embrace, extend, and extinguish competing ideas. That’s why you have 10 apps on your phone and every one of them sucks.

Decentralized Park

So what’s the play here? Do we accept that there are simply never going to be any new ideas? Do we mumble that the internet had a good run and then get back to scrolling slop on Reels? If things are going to change, it’s time for Gen Z to demand ownership of their online spaces.

Part of existing on the modern web is knowing who you actually are. Your interests, your tastes, your opinions—are they yours, or were they given to you by a machine? When you open TikTok, do you do it with purpose, or are you waiting for an algorithm to decide what slop to feed you? Did you curate your Spotify playlists with passion, or did you accept what was put on your plate?

To know the power of the algorithm is to know the face of digital ecstasy. It’s the reason you randomly pick up your phone to scroll your feeds. It’s an invisible hand shepherding you to sugary treats for your brain. The more you engage with it, the more it learns how to get you hooked. That’s what a decentralized web promises to free us from.

In a decentralized system, you decide what algorithm governs what you see, or you can elect to have none at all. What a concept! The idea of federated spaces is in its infancy (right now it’s called the Fediverse), though the greater concept of federated communications is an old one.

Your email is already decentralized. You have an email address, but you can use any software you want to send and receive email to that address. You can take your entire inbox and outbox with you to any email application, and if you decide that you no longer want to use that email address, you can create a new one and get your old email forwarded. We don’t even think about it anymore, but email is kinda magical in this regard. In fact, open standards used to be the norm until corporations started locking things down.

Decentralization isn’t limited to email. We can see it happening with social media on platforms like Mastodon, Bluesky, and Pixelfed. However, as important as it is to have ownership of those spaces, we need to be thinking of the bigger picture.

In the near-term future, our digital spaces will shrink considerably. We’re already seeing this happen with mass migration from social media forums and into group chats and Discord servers. It’s become more important than ever to have a circle of close friends and family members that you communicate with on a regular basis. We aren’t shouting our opinions to the whole internet via status updates any longer. We’re sharing memes with the group chat, and in many ways that feels right.

The more our governments crack down on privacy and online speech, the more we are driven to those local webs where we feel that we have more control, even if it’s an illusion. This is why ownership of the complete stack is necessary. As the saying goes, the heroes have to win every time, but the villain only needs to win once.

Chat Control, Age Verification, Anti-Encryption—these are the enemies of a free and open internet. Our current system is vulnerable to these laws because only a few players are in control of everything we interact with, which means that only a few of them need to cooperate for it to ruin everything for all of us. Open source hardware, decentralized software, private communications—these are the foundational tools of a free society.

We have a chance to launch a hardware and software revolution that will define the internet for decades to come, but the moment can’t wait until we’ve already been stripped of the freedoms that net-dwellers have enjoyed since the beginning. Go build something you don’t need permission to run. Self-host a Fediverse server. Tinker with a Raspberry Pi. Set up a VPN.

The best time to start owning your digital presence was 10 years ago. The second best time is when you close this tab.

As always, zack.wall@icloud.com for feedback and suggestions.