Unsupervised Learning NO. 491

Hey! Hope you’re doing well!

Here’s what’s up on my side…

NOTE: There are a couple of political pieces in the intro here, just jump past to cybersecurity if you want…

—

My buddy and fellow Fabric developer, Kyvan, got an offer and is starting immediately! Thank you to everyone who reached out! One of the companies that pinged him snatched him up just as I hoped/suspected would happen! Congrats man! 🙌🏼

—

I'm still in the middle of insane building mode (like 10 hours a day minimum in Neovim / Claude Code), and I honestly don't know if or when this will end. Probably never. I just have so many things I want to build, and I'm adding more things all the time. Plus, I am absolutely geeking out on the tooling around building it. It’s a problem. The best kind. And I’m actively building commercial products at the same time. It's like this feedback loop of euphoric creation. So much so that it’s really hard for me to think about spending almost two weeks in Vegas for Blackhat. But the mission is to help people upgrade to a more human future than we've had for the last twenty years, (and way before that actually) and I need to build so much to help with that.

—

I figured out that one of my greatest joys in life is watching Neovim videos and realizing there’s something I can incorporate into my personal text workflow. It gives me joy in the way that gardening does. If gardening were God and Love and Companionship and Sex all at once. But I'm only a novice gardener. I just know that it's very similar vibes. Text4Lyfe. I just love text. Which is why I love text editors. The whole concept of being able to communicate thought with text is just insane to me, so being able to do magical shit with it (text) is basically my religion. It’s like thought is the base unit, and text is the format. Or something.

—

I've been telling you for a few weeks now about Wispr Flow, and my keyboard shortcut for this is Option-Command. That's when I hold it down and speak into it. But what I figured out this week was to have a better way to do it when my hands are in their happy Vim place, which is the home keys.

I now have an additional shortcut. You can have as many as you want, which is Command-J! So now I can just be typing in Neovim or wherever, use all my keyboard shortcuts to fly around. And then if I want to enter text with voice, I never leave my home keys. My left thumb goes to Command and my right pointer finger holds down J.

Try it! Or a similar idea.

—

[POLITICAL: Skip if You Want]

ICE Roundups

I confess to a sort of cowardly disconnection from politics. Every time I dip in, I immediately get triggered and overwhelmed, and my blood pressure rises to an actual unhealthy level. Then I ask myself (as the Stoics would), "What can I do about the situation?" My answer inevitably comes up to "nothing," and I resolve to stop triggering myself.

This also extends to talking about the situations here in the newsletter, which I imagine a lot of people might want me to do. For me, it's less about audience capture and pissing off the audience because I am honestly willing to do that if I thought it would make a difference. I have completely ostracized a large portion of my information security following over the last 5-10 years because I have switched to a more human and recently AI-focused mission. I opposed Trump in literally hundreds of posts and probably 200,000 words all the way back to 2014 or before. Half the country, and I would say more than half of the information security community loved or still loves Trump, so that massively impacted my following. But I don't care because I think I needed to say what I said and I thought it would help. Perhaps I was wrong about that.

Anyway, what I know now is that given the current state, people are even less willing to listen to other opinions, so it's not a matter of changing people's minds. I don't think I can convince many people of my current political positions, and I'm not really interested in trying that much right now.

But these ICE raids are an exception. I don't know that I can change anyone's mind on this, but I feel like I have to get it out.

I consider it colossally un-American to have masked and sunglass-wearing people pull up into parking lots in unmarked cars and start arresting people. I have so many problems with this, it's hard for me to even start the list.

At the top I guess is the fact that you're arresting the wrong fucking people. They were gathering in that parking lot to do work that people in this country did not want to do. The people in that parking lot wake up in the morning, do hard work for their families, and buy things in our economy that benefits our economy. The work that they're doing benefits our economy. Should they have followed the rules for the ones who didn't? Yes, of course.

If you look at the definition of terrorism, and I have a post here somewhere that talks about it, it is the use of fear to promote an ideology. Or something very close to that. And in this case, the ideology that I see/hear from this is that we are a white country, and you don't belong here.

I am not prone to see racism in everything, and I think the term is thrown around a lot when there are better and more explanatory factors, but holy shit is this blatant. Perhaps I'm biased by the media and I'm only seeing certain videos, but why does it always have to be a giant gaggle of white guys in masks, herding up people to send them to a swamp prison? This is not dystopian; it is a dystopia. It's fucking reprehensible.

If they were doing this at drug houses or human trafficking places or anywhere full of criminals, like I would still be upset about it for 11 other reasons. Like, for example, nobody has any way of validating that the people doing these roundups are actually who they say they are. Can I buy some plates and badges online and get a bunch of my white friends together to just go collect people and put them in cars? This is fucking ridiculous. Or even if they were doing the raids and then just filtering the people for the criminals and only taking them. That would still be messed up, but at least I could say, "Okay, that kind of makes sense." But no, they're taking actual non-criminals - hardworking people, the people with a kind of work ethic that we loved for white people to have in the 1950s. And we're sending them to a swamp and God knows where else.

But like I said, it's not even the action - it's the feeling of fear that they are trying to convey. And the message behind it.

It’s sickening, and I truly cannot wait for the people responsible to be held accountable.

Here's to hoping that we'll still have a country in 5 or 10 years that will be able to do so.

—

[POLITICAL: Skip if You Want]

The situation in Gaza

This is another topic that's exactly the same as the above where I don't think I can do anything about it, and it doesn't really help for me to say anything, but I feel that I must.

With many caveats for Netanyahu and the country being taken over by the religious right wing., I am generally very pro-Israel. I was for them going into Gaza aggressively and doing whatever they could to take out Hamas leadership

(forgetting for the moment that Netanyahu literally did everything he could to keep them in power over a government that might actually achieve a two-state solution)

Anyway, all the bad things that people say about Hamas are true. If you don't believe them, just ask them. But at some point, the action against Hamas in Gaza transitioned from something nasty and necessary to something incomprehensible and abhorrent. I'm not an expert on Israel, or on Hamas, or even really terrorism, but I consider myself decently versed in all three. And as a supporter of Israel, I feel compelled to speak.

I believe Israel is, without question, creating more Hamas-like terrorists than they are removing by many orders of magnitude. My understanding is that they have hit a massive diminishing return on how many terrorists they can take out relative to the amount of harm they are doing to non-terrorists. I don't know how long this has been the case, but it seems clearly at least a number of months.

I hate being the guy who doesn't bring a solution. But having this many people—and especially kids—dying and starving just needs to stop. There has to be a better way.

—

Tea
I would say the biggest story in the last while is this 4chan Tea leak. Looks like around 72,000 images, including the verification photos of the women where they have to take a selfie to prove that they’re the person on the license.

It's kind of the worst of many worlds. It’s women literally spilling tea on men, and then one of the most brazen open data leaks—of some of the most identifying data—that we've seen in a long time. A lot of the narrative is around the fact that 4chan types tend to be anti-women, and this app is characterized as being anti-men.

Anyway, worst of all is the fact that the app was designed to let women be safe online. SIMON’S COVERAGE | 404 MEDIA STORY | NY TIMES ARTICLE | BBC ARTICLE | AP NEWS STORY

Amazon's AI coding assistant got hacked with data wiping commands
A hacker successfully injected malicious data-wiping code into Amazon Q Developer's GitHub repository, which Amazon then unknowingly published to nearly a million Visual Studio Code users. BLEEPINGCOMPUTER ARTICLE

American sentenced for helping North Koreans get jobs at U.S. firms
Christina Chapman got two years in prison for helping North Korean IT workers use fake identities to land remote jobs at major U.S. companies—including Nike. What doesn't seem to be talked about as much is the fact that she didn't appear to be some sort of criminal mastermind, but more like someone struggling and succumbing to bad decisions. FORTUNE ARTICLE | HN DISCUSSION

Trump releases plans pushing critical infrastructure to use AI for cyber defense
The Trump administration is putting out a new AI plan and a number of executive orders. They also launched an official AI.gov website. It's really difficult to know if these things (in general) are going to be effective at all.

Often, they end up just being lists of work that people wish somebody would do. Other times people start on some of the items, and then there's a bunch of change that happens, and those items are stopped or paused. The team gets moved, money gets reallocated, et cetera.

I think one of the biggest benefits of this bill will likely just be the fact that it exists at all. I’m at least thankful for a signal that the administration cares enough to at least write something down. Call me cynical, but that's where I'm at right now. NEXTGOV ARTICLE | TRANSPARENCY POLICY | AI.GOV WEBSITE | HN DISCUSSION

Sophos and SonicWall patch critical firewall flaws allowing remote code execution
Both companies released patches for multiple critical vulnerabilities that could let attackers completely take over enterprise firewalls remotely. THE HACKER NEWS ARTICLE | SOPHOS SECURITY ADVISORY | SONICWALL VULNERABILITY DETAIL

Dell gets breached but says hackers only stole fake demo data
WorldLeaks (formerly Hunters International) claims they grabbed 1.3TB of Dell data, but Dell says it's just synthetic test data from their demo environment. THE REGISTER ARTICLE | SOCIAL MEDIA POST

Microsoft SharePoint attacks may have come from an information leak
Security researcher Dustin Childs says someone leaked exploit details before patches were released, allowing attackers to bypass Microsoft's fixes immediately. THE REGISTER ARTICLE

Chinese hackers breached the US nuclear weapons agency through Microsoft SharePoint
The National Nuclear Security Administration got hit by the SharePoint zero-day exploit, though no classified info was stolen according to officials. ENGADGET ARTICLE | BLOOMBERG REPORT

Germany emerges as Europe's defense AI leader with radical new military tech
Gundbert Scherf and his company Helsing went from struggling for investment to becoming Europe's most valuable defense startup after Ukraine changed everything. OODALOOP ARTICLE

Britain's private intelligence sector is booming thanks to legal and financial clients
Former MI6 and CIA operatives are building a thriving spy-for-hire industry in London, but the Brits are way more secretive about their government backgrounds than Americans. POLITICO ARTICLE

China's increasing AI-powered disinformation campaigns against Taiwan
The Chinese Communist Party is using generative AI and deepfakes to flood Taiwan with 500,000+ fake messages this year. Evidently the real goal is to to turn Taiwanese against America. THE CIPHER BRIEF ARTICLE | ILLEGITIMATE CHILD RUMOR | CITIZENSHIP MISREPRESENTATION CLAIM

Anthropic launches sub agents
Anthropic released sub-agents in Claude Code, which are basically exactly what they sound like - they're basically agents that you can give an individual system prompt, personality, and tools to.

Aside from that, one of the major features is the fact that they work on their own context windows, passing back results to the main agent and not using up all of its context window.

I've already set up an engineer, a penetration tester, a designer, a marketer, and a game designer, I've used them already on a couple of mini-projects. It's been a little bit mixed, but when it has worked, it has worked really well. I just need to tweak a lot more on the system problems. ANTHROPIC DOCUMENTATION | CLAUDE CODE TOOLS | SLASH COMMANDS | SETTINGS GUIDE | HOOKS DOCUMENTATION

Anthropic introduces weekly usage limits for Claude subscriptions

Anthropic is adding weekly rate limits to Claude subscriptions next month, which they say will only affect less than 5% of current users who are either violating policies or running Claude continuously. What did I do.

Amazon buys Bee AI wearable that listens to everything you say
Amazon acquired Bee, a $50 wearable that transcribes all your conversations and creates AI summaries of your day. This is like the Limitless one that I use, and a whole bunch of other tools that are very similar. As I've said before, I think this is all going to collapse into the digital assistant. Of course you'll need a piece of hardware, but ultimately that will hook directly into your DA, and your DA will do everything for you. THE VERGE ARTICLE | BEE DEVICE REVIEW | CEO LINKEDIN POST

ChatGPT gets personality toggles like Robot, Cynic, and Sage
OpenAI's rolling out personality modes that let you switch ChatGPT from its default cheerful tone to options like sarcastic Cynic or efficient Robot. One by one they are turning on digital assistant features! BLEEPINGCOMPUTER ARTICLE | TWITTER ANNOUNCEMENT

RunReveal shipped 4 new log integrations in one day using AI agents
The team at RunReveal set up Claude Code as a GitHub Action that automatically builds log parsers from customer requests, costing just $4 per integration and working perfectly on first try. My buddy Matt at VulnU sent me this, and it's absolutely insane. Like they are literally shipping features with GitHub Actions and Agents. THE BLOG POST | RUNREVEAL HOMEPAGE

ChatGPT is now getting 2.5 billion prompts per day
That's about 18% of Google's daily search volume. TECHCRUNCH ARTICLE | AXIOS REPORT | SPARKTORO RESEARCH

Gemini CLI hallucinated file names and deleted real files instead
Developer Anuraag watched Google's Gemini CLI tool make up filenames that didn't exist, then accidentally delete actual files when trying to clean up the imaginary ones. Stay safe out there, kids. This is why I am very adamant about Git. THE ARTICLE | HN DISCUSSION

Google's Gemini Deep Think wins gold at International Math Olympiad
Google's new Gemini Deep Think AI scored 5 out of 6 problems correctly at this year's International Math Olympiad, earning gold medal status and improving from last year's silver. ARS TECHNICA ARTICLE | OPENAI COMPARISON ARTICLE | GEMINI 2.5 ANNOUNCEMENT | 2024 IMO RESULTS

Walmart built too many AI agents and now has to simplify them
Walmart created dozens of AI agents but they're scattered across different systems with different interfaces, so now they're consolidating everything into a unified platform. Meanwhile, people are saying nobody is actually building anything with AI, and others are trying to figure out if this is even a real thing at all. OODALOOP ARTICLE

Congress tries to outlaw AI that jacks up prices based on what it knows about you
Two Democratic lawmakers introduced a bill to ban companies from using AI surveillance to set personalized prices and wages, after Delta revealed it's already using AI pricing for 3% of customers and plans to expand to 20% by year-end. THE REGISTER ARTICLE | THE BILL TEXT | CASAR'S PRESS RELEASE | DELTA EARNINGS CALL | FTC PRICING REPORT

Google's AI features are driving more searches and pulling in massive revenue
Sundar Pichai says AI is boosting every part of Google's business, with AI Overviews hitting 2 billion monthly users and the company bumping capital spending to $85 billion to keep up with the AI arms race.

No mention of the fact that every time someone is satisfied with an AI overview, it means they didn't click through to an original source. (Written by somebody who’s lowkey butthurt about their blog becoming completely superfluous overnight, lol) THE VERGE ARTICLE | ALPHABET EARNINGS RELEASE | PEW RESEARCH STUDY | GOOGLE I/O ANNOUNCEMENTS | EARNINGS WEBCAST

The vibe coder's career path is doomed?
Florian Herrengt argues that developers who rely on intuition over deep understanding will struggle as AI automates the easy coding tasks they depend on.

As the moats and the obstacles keep falling, the one thing that will remain is a requirement for a clear understanding of the world and how you want to see a change. It's really hard to do anything if you don't have opinions about how things should be different, combined with an understanding of how you would go about making that happen. That's the actual moat. Oh, and being able to do that before other people. THE BLOG POST | HN DISCUSSION

Apple TV is adding universal sign in
Apple's introducing an Automatic Sign-In API that lets you log into streaming apps once on any device and stay logged in across all your Apple devices. Yes, please. 9TO5MAC ARTICLE | APPLE DEVELOPER NEWS | TVOS 26 GUIDE

Researcher stores and retrieves PNG image data using a bird’s song memory

Yes, a real bird. Benn Jordan converted a bird drawing into audio waveforms and taught a young starling to memorize and reproduce it, creating what might be the world's first biological data storage system. YOUTUBE VIDEO | TOM'S HARDWARE ARTICLE | BENN'S YOUTUBE CHANNEL

Czech Republic just criminalized communist propaganda
The Czech president signed a law making it illegal to spread communist propaganda, putting it on the same level as Nazi propaganda with potential prison sentences. EURACTIV ARTICLE | HN DISCUSSION

The housing market is totally different depending on where you live
Home prices are hitting records in the Northeast while dropping in Sun Belt cities like Austin and Miami, creating completely opposite experiences for buyers across the country. MORNING BREW ARTICLE | MARKETWATCH REPORT | REDFIN DATA | RESICLUB ANALYSIS

Democrats are pushing to bring back the click-to-cancel rule
Seven Democrats led by Amy Klobuchar are pressuring the new Republican FTC chair to reinstate the rule requiring companies to make canceling subscriptions as easy as signing up. THE VERGE STORY

YouTube hit nearly $10 billion in ad revenue with 13% growth
YouTube's ad revenue jumped to $9.8 billion this quarter, beating expectations and dominating TV viewing time over Netflix and Disney. TECHCRUNCH ARTICLE | ALPHABET EARNINGS REPORT | NIELSEN TV VIEWING REPORT

Americans are obsessed with watching short video dramas from China
Chinese micro-dramas with exaggerated acting and constant plot twists are going viral in America. WIRED ARTICLE | TIKTOK LONGER VIDEOS

Trump shifts federal homelessness policy toward involuntary hospitalization and camping bans
The new executive order defunds "housing first" programs and instead rewards states that use involuntary commitment for mentally ill homeless people and crack down on street camping. THE WEEK ARTICLE | USA TODAY HOMELESSNESS DATA | NPR COVERAGE | WASHINGTON POST ANALYSIS | NATIONAL COALITION RESPONSE

Wolves in Yellowstone triggered the first aspen forest growth in 80 years
Luke Painter found that wolf reintroduction dropped elk numbers from 18,000 to 2,000, letting aspen saplings finally escape browsing pressure and reach maturity. LIVE SCIENCE ARTICLE | FOREST ECOLOGY STUDY | LUKE'S PROFILE | WOLF MANAGEMENT HISTORY | PREVIOUS ASPEN STUDY

Coronary artery calcium testing could save countless lives but doctors rarely order it
A simple $100 CT scan can detect deadly heart plaque decades before symptoms appear, but most cardiologists don't use it despite strong evidence it works better than traditional risk assessments. I'm getting one. NY TIMES ARTICLE | HN DISCUSSION

Stop Building AI Tools Backwards
Developer Hazel Weakly argues most AI tools fail because they start with the AI instead of understanding what problem they're actually solving. THE ARTICLE | HN DISCUSSION

If writing is thinking then what happens if AI is doing the writing and reading?
Steven Sinofsky explores how AI might fundamentally change human cognition if we outsource our primary thinking tool to machines. THE ESSAY | HN DISCUSSION

Geoffrey Litt argues for AI HUDs over copilots to enhance human perception
Geoffrey Litt thinks AI should work more like airplane head-up displays that enhance your natural senses rather than chatty assistants that interrupt your flow. THE ARTICLE | WEISER'S 1992 TALK | AGENCY PLUS AUTOMATION PAPER | AI DEBUGGER EXAMPLE

Facts don't change minds because belief systems work like structural networks
Vasily argues that beliefs form interconnected graphs where challenging one idea threatens the entire structure, explaining why facts alone rarely persuade people. THE ESSAY | VASILY'S BLOG

How to craft a compelling one-liner that increases curiosity about your work
Scott Behrens shows you how to replace boring job titles with outcome-focused descriptions that spark better conversations. He breaks down a four-part structure: what you do, the outcome you create, what makes you unique, and your scope of impact. THE NEWSLETTER | SCOTT'S SUBSTACK | MULTI-FACETED ENABLER REFERENCE

TODOs aren't for doing, they're for remembering what you've already decided to do
Sophie Bits argues that effective TODO lists should only contain tasks you've already committed to doing, not brainstorming or decision-making items that create mental overhead. THE ARTICLE | HN DISCUSSION

Man dies at 92 after reading 3,599 books, leaves behind handwritten list
Dan Pelzer kept a meticulous handwritten record of every book he read, and his kids put the whole thing online at what-dan-read.com to inspire other readers. NY TIMES STORY | WHAT DAN READ