Related to your interests - Week of March 2nd, 2025
No time for coming up with a strapline, otherwise I'll never actually click send.
The Links
Claude, without me asking, rewrote many of my original description below. Enjoy!
Pentagon Leverages AI in Iran Strikes Amid Feud With Anthropic - One artillery unit doing the work of 2,000 staff with a team of just 20 people. Anthropic didn’t want their model used for this; the Pentagon used it anyway. The future is here, it’s just unevenly distributed between PR statements and actual weapons targeting.
Anthropic’s Pentagon Feud Accelerates Push Into Consumer Market - Claude’s free active users grew 60%+ and daily signups grew 4x since the start of the year. Apparently nothing boosts consumer signups like a military controversy.
Pentagon’s Anthropic Designation Won’t Survive First Contact with Legal System - “The legal problems are so glaring, in fact, that a cynical possibility suggests itself: The administration knows this won’t survive judicial review and is doing it anyway, so that when they inevitably lose, they can still claim to have gone hard against Anthropic. This is designation as political theater: a show of force that was never meant to stick.”
U.S.-Iran war threatens Gulf AI infrastructure as both data chokepoints close - Thrilling times for IT threat modelers: billions of dollars in U.S. technology infrastructure, and trillions more in planned investment, now depend on fiber-optic cables running through war zones. Or, they could just bomb the data centers directly.
Manipulating AI memory for profit: The rise of AI Recommendation Poisoning - Attackers manipulating what your AI “remembers” to steer its recommendations. A new attack surface nobody really planned for when they bolted memory onto LLMs. This is done with URL-driven prompt injection.
Ultimate prompting guide for Nano Banana - From Google, so it’s not web garbage trying to sell you a “guide.”
How Oren personalizes his AI stuff - His CLAUDE.md setup, etc.
Skilled Agents - Bill goes over some more AI rig personalization.
Creative Work in an Age of Digital Production - “MrBeast’s great strength as a contemporary creator is that he has no ambition beyond repeating a pattern. He’s a machine-listener.” There’s something a little missing here though — his execution of the pattern is the thing. It’s a genuine evolution of reality TV and game shows. Plus, really good YouTube thumbnails.
🤖 Reversible men and Lipský’s Happy End - The protagonist Bedřich Frydrych is “born” via guillotine reattaching his head, his murdered wife is reassembled from pieces in a bathtub, and the axe is pulled from her forehead to revive her. Frydrych’s cheerful voiceover reframes every atrocity as a positive domestic development.
Meta Ray-Ban smart glasses footage reviewed by contractors -
Wastebook
“A standard confused NINO account.” Russel
Sometimes, CEOs are not reliable narrators.
If Claude created images, it’d be a full replacement of the other AIs.
“six-page narratives in seconds.” This changes everything.
“Stick with ‘meh’ because it’s safe.” WIP.
I asked Nano Banana to make a slide for me, and it made the slide, but as if it were a photograph I’d taken at a talk. It included being framed poorly, cutting off half the title and crooked. Amazing!
“a hotel-led, mixed use transformation.” He got a leaflet.
That moment when you’re watching some streaming service show and you’re like “fuck, I’m two seasons in and I just realized this is just a well produced soap opera.”
Some people survive by embracing the absurdity of the BigCo, sort of as a TV show they watch from time to time. Doing that would make you cynical, but the other part is working on what you think matters in the meantime.
ICYMI
When to use AI for writing, and when it’s totally acceptable
I’ve been working on some faux game cards to use a schwag ar work. They were finally released in the wild. Here’s a picture of two. I’m making a series two for Spring I/O.
Software Defined Talk Episode 562: Bureaucracy: Still Unsolved - This week. we discuss Claude Code’s momentum, Cursor’s identity crisis, and the SDLC’s uncertain future. Plus, Coté finally explains how Markdown is destroying the economy.
Software Defined Interviews Episode 121: Art Degrees, Sun Microsystems, and How Kubernetes Scales Contributions, with Josh Berkus - Whitney and Coté discuss with Josh Berkus (Red Hat, Kubernetes contributor) how liberal and fine arts degrees (philosophy, photography, sculpture, pottery) apply to tech careers. Berkus details how early hardware experience influenced his database performance work, noting hardware’s renewed relevance with AI and multi-arch computing. The conversation covers Sun Microsystems’ 1990s internet role, internal politics, and its MySQL/Postgres strategy. They examine open source’s shift from end-user to vendor-driven models, foundations’ roles, and contributor incentives. Berkus describes Kubernetes release processes, contributor-experience programs, and its resilience to low-quality AI contributions.
Logoff
I’ve started to use Claude Core for non-programming agentic stuff. Once you set up the memory system (just storing context in various ways), it is, indeed, amazing.