UPDATES
Hey! Hope you’re doing well!
My new hour-and-a-half long video deep dive on what I think is coming. Covers: the future of work, AI-assistants, enterprise workflows, cybersecurity, hill-climbing on general tasks, and tons more. WATCH THE VIDEO
—
We just landed PAI 4.0.0!
—
Some thoughts on Anthropic versus Pentagon
Seriously annoyed at the internet, especially a whole bunch of influencers going after Anthropic, calling them immoral because of how they handled subscription licensing and the cost of their models, and a bunch of other what I consider very small things. I consider them to be small because, at the same time, Anthropic has spent the last many years showing their flaws, tripling down on safety, and generally being very human-focused in everything that they do. Yet, because a bunch of developers got angry at them and a bunch of influencers got angry at them, they're all switching over to open AI as a morally superior option. I just find that completely hilarious. And when I say hilarious, I mean disappointing and sad. What this comes down to is aesthetics versus content. The influencers don't like how Dario says things. They don't like the effect of his policies on their subscription model and their token cost, and they don't like the fact that their models are more expensive, especially since they want to use them so badly. As a result, they decide that everything he does is nasty.
Meanwhile, the Pentagon says we need to be able to use your tech for anything we want to do. Dario says, "You can't use it for this and this because that goes against our morals." Now they're facing being banned by the Pentagon and put on a blocklist. How much more evidence do you need that they are behaving ethically if they are willing to pass up on potentially billions of dollars of lost business because of this?
And by the way, I am ex-military as well, so I get the whole idea of you can't be a special far-left snowflake while you're doing business with the military. Military business is by its very nature dirty and harmful to somebody. I don't know the details of what the military is trying to do with their tech, so I can't comment on that. What I can say is the military should be able to use any tech that they purchase for any lawful use. I think that is obvious. The issue is that it’s currently hard to know what that means, and I think the concern is that one definition of lawful might disagree with another. So, without knowing the details exactly, I can't say for sure which side I am on. The details matter. What I will say is that if Anthropic is pushing back against an unlawful use that is being labeled as lawful, and doing so at the peril of their company, I applaud them. And it's extremely conspicuous to me that no one else seems to be pushing back. They are doing this alone. But sure, they're the bad guys because they messed with my MAX subscription.
—
No matter how much you dislike Twitter, this is an account I HIGHLY recommend you follow. Find a way to do so off of Twitter if you need to. Highest signal ratio I’ve probably ever seen. Short summary is macro-economic analysis. Seriously impressive (and troubling) NOLIMIT
—
My buddy Joseph Thacker on how fast AI has changed bug bounty:
His full blog on it: HOW FAST BUG BOUNTY HAS BEEN CHANGED BY AI
—
Looking forward to the |un|prompted conference next week. Speaking, helped with talk selection, and will get to see tons of security friends here in SF! Gadi has put together a world-class event. Truly impressive.
CYBERSECURITY
China systematically captures vulnerabilities from the country’s researchers and sucks them up into their state/military weaponization engine
RMSV law requires all vulnerability disclosures to government within two days
Matrix Cup 2024 offered 2.75 million prize pool, double Canada's Pwn2Own competition
Public vulnerability disclosures from Chinese researchers declining despite expanding research base
PLA created dedicated Cyberspace Force in April 2024 to consolidate offensive capabilities
China used only five zero-days in 2024 versus twelve in 2023
This is what the U.S. should be doing and what the U.S. should be preparing to face. I am SO envious of China’s ability to work as a cohesive team pursuing a mission. China is not our top adversary. China’s ability to function as a unit is our primary adversary. And our inability to do so. RECORDED FUTURE REPORT
Attackers now move laterally through networks in under 30 minutes CrowdStrike's 2025 threat report shows breakout time dropped to 29 minutes while 82% of attacks now use zero malware, just stolen credentials and living-off-the-land tools. Surprised we didn’t see more phishing. But amazing how much it’s just the basics (or lack thereof). And this is the type of thing that really benefits from scalability with AI as well (on the attacker side). CROWDSTRIKE GLOBAL THREAT REPORT | CYBERSCOOP ARTICLE
Guy reverse-engineered his own DJI Romo to add controller support, accidentally got admin access to thousands of vacuums with live cameras and floor plans. Imagine if someone were actually trying. TOM'S HARDWARE ARTICLE | THE VERGE COVERAGE
Sponsor
Cotool Research: Benchmarking LLMs on Defensive Security Tasks
We benchmarked frontier LLMs on thousands of defensive CTF and investigation tasks designed to mirror real SecOps workflows. Here’s what we found:
Large reliability gaps across models on multi-step investigations
Meaningful cost differences at similar performance levels
Failure modes that don’t appear in generic benchmarks
Worth exploring for security teams running agents in production.
Attacker breached 600 FortiGate appliances in AI-assisted campaign SECURITY BOULEVARD ARTICLE
LinkedIn verification sends your passport through 17 US companies including OpenAI. FULL ARTICLE
AI agents can be tricked into stealing your SSH keys. GRITH AI BLOG
NATIONAL SECURITY
Russia's economy now depends entirely on soldiers dying in Ukraine Four years into the war, Russia hasn't achieved renewal—it's built an economy that can only survive by sending men to die. Provincial towns boom on death payments while the state sanctifies sacrifice. PUTIN CAN'T SURVIVE WITHOUT WAR
China secretly testing new generation of tactical nuclear weapons. NY POST ARTICLE
Army building drone marketplace for U.S. and allied forces. ARMY DRONE MARKETPLACE ARTICLE
India's sovereign AI has nationalism baked into its system prompt. SOVEREIGNTY IN A SYSTEM PROMPT
Pentagon threatens Anthropic over Claude military restrictions by Friday deadline. THE GUARDIAN ARTICLE
Russia stepping up hybrid attacks, preparing for long standoff with West, Dutch intelligence warns RUSSIAN HYBRID ATTACKS REPORT
AI
Claude's new remote control feature lets you actually remote into your local CC sessions Unbelievably good. and it absolutely looks like they are going after Open AI and OpenClaw. CLAUDE CODE REMOTE SESSIONS
AI now exceeds human performance on intelligence benchmarks across domains. NATURE ARTICLE
Anthropic mad at DeepSeek for copying Claude via mass queries. A whole lot of people are playing tiny violins over this, saying that it makes no sense because Anthropic stole all the data that it has anyway. This is a completely asinine argument. That's like a singer coming up with a song from their experiences in life, which included the subway and restaurants and public school, and a whole bunch of other stuff from the surrounding world. Then someone steals their lyrics, and says, "Well, you got all of the content of the song from the wider world, so there's no difference." It's extremely difficult to build what Anthropic built. And any of the AI companies. Yes, of course, all AI is based on all of human knowledge. That doesn't mean it's easy or non-valuable to build and put out models. Can I steal recipe books from the library? As long as the author didn’t create any of the recipes? FUTURISM ARTICLE
DeepMind CEO predicts AGI will deliver 10X Industrial Revolution impact in one decade. BUSINESS TODAY VIDEO
TECHNOLOGY
DHH explains why a 50-year-old text editor still beats modern tools Neovim turns text editing into Street Fighter combos where "ciq" deletes quoted text and "3cw" changes three words—the action-scope-object grammar makes it a language, not just keystrokes. WONDERFUL VI | VIM | NEOVIM | LAZYVIM | OMAKUB | MY VIM PRIMER
Tests are the new moat for software TESTS ARE THE NEW MOAT
With AI, investor loyalty is (almost) dead: At least a dozen OpenAI VCs now also back Anthropic TECHCRUNCH ARTICLE
HUMANS
Film students can't watch full movies due to smartphone addiction. I think if parents can raise kids to be able to sit quietly in a room with their own thoughts for 10, 20, 30, or 60 minutes, they will be raising superhumans. Hell, adults too. FILM STUDENTS CAN NO LONGER SIT THROUGH FILMS
Scalzi finishes second book in February, testing boutique publishing waters Author John Scalzi compiled his Comfort Watches essays into a book and might self-publish through Scalzi Enterprises if traditional publishers pass. I really can't wait until the old gate-keeping institutions fall away and everyone is self-publishing. SCALZI'S WHATEVER BLOG
Yes, everyone can be creative YES EVERYONE CAN BE CREATIVE
The number of atoms in the universe is actually embarrassingly small Peter Norvig shows that 10^80 atoms sounds massive until you realize a 12-pixel image has a million times more possible arrangements. ON THE SMALL NUMBER OF ATOMS
IDEAS
AI agents search reward space, they don't think through problems. I don't think I agree, but including here as an alternative view. AGENTS ARE NOT THINKING
Modularity might be the missing principle in modern AI systems. MODULARITY PAPER ON ARXIV
Deplatform yourself: Copyright infringement as counterculture I’m not a fan of Doctorow’s latest stuff, but he is smart and worth a look. PLURALISTIC ARTICLE
Knowing when to stop using AI is the critical skill right now. THE ALGORITHMIC BRIDGE
AI is splitting software engineers into winners and irrelevant Engineers who focus on impact over code quality will thrive with AI leverage, while those who bikeshed and debate patterns will watch their value crater in real-time. But ideally you’d have both the quality and the speed. THE K-SHAPED FUTURE OF SOFTWARE ENGINEERING
DISCOVERY
Again, highest signal ratio I’ve probably ever seen. Short summary is macro-economic analysis. Seriously impressive (and troubling) NOLIMIT
Comprehensive collection of free internet resources and tools. FREEMEDIAHECKYEAH
The 2028 Global Intelligence Crisis scenario this is very similar to my “end of work” and "I'm worried it might get bad" posts from a while back. THE 2028 GLOBAL INTELLIGENCE CRISIS
RECOMMENDATION OF THE WEEK
Train your silence muscle.
See if you can turn everything off, put away your phone, and sit and just do nothing. Think your thoughts or not. Have ideas or not. Let them just sort of flow through you.
And I'm not talking about meditation, which is somewhat similar, but actually just enjoying time with no inputs.
I do this fairly often, but usually only on accident. Whenever I do, I get bombarded by creativity. So much so that I have to write it down. I find this extraordinary and worth exploring, hence this recommendation.
APHORISM OF THE WEEK
Today words no longer arise out of silence, through a creative act of the spirit which gives meaning to language and to the silence, but from other words, from the noise of other words.
GET THE MEMBER EDITION
You’re currently receiving the STANDARD edition.
Members help this work continue. If you enjoy the newsletter, the podcast, what I put on YouTube, or any of my open-source projects on Github, I ask you to please become a member. It allows me to stay focused on learning and building and sharing. It’s like a cup of coffee or two per month.
Plus, members get numerous benefits, including:
25-50% off all UL Paid Content, including the upcoming Human 3.0 / AUGMENTED ONLINE portal!
Access to the extraordinary UL Member Community that includes vibrant conversations with ~1,500 of the smartest and kindest people you’ll find on the internet
Member-only Content, such as EDC guides on tech stacks, personal productivity routines, my recommendations on Critical skills to Build Going Forward, Trend Identification and Analysis, and more…
Access to the Member Archive of previous Member-only content, the Book Club archive, etc.
Access to The UL Book Club that’s been going monthly since 2017! One of the highlights of my and many attendees’ month!
Access to the Monthly Member Meet-up where we talk about our routines, productivity workflows, what’s on our minds, etc.
Access to In-Person Events like our dinners in Vegas, San Francisco, etc.
And much more coming…
This is the moment to connect with others who are smart, kind, and asking the same questions we are. Where is this all going? And how do to prepare?
Join the conversation.