AsymmetricalCollage / AC is on Nostr

"56395162febb6b491f60afd601e25e164c39f190a62ef643312ae3e7a828761ee3067a741331f6823946db37e2888de0cb8332c1867af237d3a781180392004b"

Last Notes

2026-04-11 15:13:48 GMT

- reply

Read the same Karpathy post and had the same inclination - build my own. Ended up testing yours first because you're not some random internet dude. One thing I keep coming back to is the token-intensive nature of the research. Groups working on the same or adjacent topics end up duplicating these costs independently. There's probably something interesting in pooling raw source piles across trusted peers. Shared resources, independent Q&A and output processing. Curious if you've thought about the pooling angle.

2026-04-11 13:54:22 GMT

The skills framework was built for a world where humans do the execution.
Hard skills: technical. Writing code, reading a balance sheet, modeling risk. Soft skills: interpersonal. Communication, leadership, judgment under pressure. It's a clean taxonomy, and for most of the twentieth century it held.
AI breaks it at the execution layer. It writes code. It reads documents. It models faster than any analyst. The hard skills most people spent years acquiring are now table stakes — not because they stopped mattering, but because the machine can cover them.
What remains isn't a soft skill. It's something the old taxonomy didn't name, because humans had no choice but to execute themselves.
The new differentiator is knowing what to ask. Whether the output is right. What problem is actually worth solving. When to override. When to trust. These aren't soft skills — they're not interpersonal, they're not about managing people or reading a room. They're something else: a capacity to orient before the machine starts moving.
I built my entire research and publishing system with Claude Code. I'm not a developer. I built it because I could frame the problem clearly, evaluate the output honestly, and iterate from failure. The machine executed. I supplied the judgment.
That's the category. Not hard, not soft. Call it orientation: knowing what good looks like before the machine starts.
People are racing to learn skills the machine already possesses. The more useful question is what you build that it doesn't.
Adapted from Asymmetrical Collage.
https://www.asymmetricalcollage.com/articles/2026-04-11-what-remains/

2026-04-10 06:41:03 GMT

Reports this week: Iran is demanding tolls on tankers transiting the Strait of Hormuz. Acceptable payment: Bitcoin, or yuan. Not dollars.
This is not a Bitcoin adoption story. It is a settlement story.
Iran cannot clear in dollars. SWIFT is not an option. When the institution is the threat model, you find a settlement layer that doesn't require the institution's permission. Bitcoin qualifies. So does yuan. The common feature is not ideology. It is neutrality.
The Strait carries one-fifth of global oil trade. There is no alternative route. If the toll holds, every barrel in transit is a live proof-of-concept for neutral settlement.
Monetary sovereignty advocates have described this scenario for a decade. McKinsey put it in numbers last month: gold above $5,000, 60% of CFOs citing geopolitical risk as their primary concern, trade settlement assumptions under legal challenge.
The argument for neutral money is most compelling when the alternative is unavailable. The world is making the case faster than any advocate could.
Adapted from Asymmetrical Collage.
https://www.asymmetricalcollage.com/articles/2026-04-10-neutral-ground/

2026-04-09 07:08:02 GMT

Nuclear weapons were a state secret before they were anything else. GPS was military for twenty years before it was a phone app. Strong cryptography was export-controlled before it was a browser standard.
Technology powerful enough to reshape power doesn't reach the public first.
Anthropic just ran the same playbook. Claude Mythos Preview autonomously found zero-day vulnerabilities in every major operating system and browser. It chained Linux kernel vulnerabilities to escalate to root. It found a 27-year-old flaw in OpenBSD and a 16-year-old bug in FFmpeg that had survived five million automated tests. Zero-days like these are routinely exploited for months or years before they surface. The public never knows when.
The private list of testers: AWS, Apple, Cisco, CrowdStrike, Google, JPMorganChase, Linux Foundation, Microsoft, NVIDIA, Palo Alto Networks. All American. Not publicly available.
Controlled deployment is defensible. If attackers have it before defenders can prepare, the harm is real. The companies in this coalition run infrastructure the public depends on. Anthropic also has a direct interest: an exploit traced to Mythos would be commercially catastrophic. Both are true. They are not the same argument.
Every participant is American. Europe, Asia, South America, Africa — no seat. Not designed for the world. Designed for the current guard.
The question Glasswing raises but doesn't answer: if this is what gets disclosed, what exists that doesn't? Who holds it? Under what governance? What protects the rest of us from whoever runs the version we never hear about?
The precedent exists. Bitcoin was built on the premise that the institution was the threat model, not the technology. That produced a different architecture: global mailing list, no coalition, no national preference, no gatekeepers. Public first.
Open source cryptography protected the public from surveillance capitalism faster than legislation ever could. The same race is now underway in AI. The heroes of security and privacy who understand what is at stake need to deliver open models to the general public at the same speed private companies are closing access. The public's defense cannot depend on a coalition it was never invited to join.
Adapted from Asymmetrical Collage.
https://www.asymmetricalcollage.com/articles/2026-04-09-the-private-list/

2026-04-08 07:58:02 GMT

Every payment you make is a choice. Which rails. Which intermediary. How much visibility you keep.
You may have thought carefully about this. Most people haven't.
Now you're delegating that choice to an agent.
HTTP 402, the "Payment Required" status code, was reserved in the early web's protocol stack, waiting for digital money to arrive. It waited thirty years. AI agents are filling it now. There are already 15,000 paid API endpoints built for agents to discover and transact with. Three competing protocols fighting for the rail. The agents are already paying.
The question isn't which protocol wins. The question is: who authorized the agent?
When you pay, you consent. When an agent pays on your behalf, the consent is buried in configuration files and developer defaults. You approved the agent. You did not approve every payment it makes.
Sovereignty isn't just about choosing the right rails. It's about knowing what runs on them, and who decides that when you are not watching.
The defaults are being set right now. Ask the question before someone else answers it for you.
Adapted from Asymmetrical Collage.
https://www.asymmetricalcollage.com/articles/2026-04-08-who-authorized-the-agent/

2026-04-07 06:43:34 GMT

A customer taps their card. The merchant receives roughly 97 cents.
The rest goes to the parties who own the pipes: interchange to the issuer, scheme fees to the network, a margin to the acquirer, a cut to the processor. The customer pays to rent infrastructure they did not choose.
Those same pipes can say no. A merchant delisted. An account frozen. A transaction blocked. Payment and permission flow through the same pipes.
Merchants accept this because the alternative is commercially painful. Issuers fund cardholder rewards from interchange, which builds consumer demand, which makes refusal costly. Acquirers lock in merchants with terminals, working capital, and favorable rates. By the time a merchant questions the arrangement, switching costs have compounded for years.
Bitcoin fees work differently. The sender pays miners, who contribute hash power to secure the network. The fee is not pipe rental. It is compensation for the security that makes the transaction final. And the system is designed to sustain this: as block subsidies decrease with each halving, transaction fees are built to take over as the primary miner incentive. More use, more fees, more security. The loop is intentional.
In fiat, the fee is rent. In Bitcoin, it is an investment in security.
Originally published at:
https://www.asymmetricalcollage.com/articles/2026-04-07-rented-rails/

2026-04-06 13:30:03 GMT

Two years of AI discourse has been dominated by prompt engineering. Craft a better question, get a better answer. There's something true in it. But it's not where the real leverage is.
What actually moved the needle wasn't how I phrased questions. It was the context that existed before any question was asked. The more accurately the AI knew who I was, what I was working on, what I already knew — the better the output. Not because my prompts improved. Because the environment did.
This distinction matters enough to name. A prompt engineer optimizes the question. An AI Environmentalist designs the conditions in which good thinking grows. An instruction tells the AI what to do right now. An environment tells it who you are, how you think, what you're working on, and what you value — persistently, across every session, without re-explanation.
I built this on Obsidian and Claude Code. The vault is organized around states, not topics. Notes move through stages of development. There's a governing document that loads at every session start — not a config file, a teaching document. Memory persists across sessions in two layers. The AI gets smarter about my life over time. Not because the model improved. Because the environment contains more accurate information.
The whole thing runs locally. My notes are markdown files on my machine. No proprietary format, no subscription holding my data hostage, no platform that changes its terms. The system is mine and can remain that way. That's not a technical footnote — it's the point.
Before building this: a blog started four years ago. Struggled weekly, published 12 posts, stopped. Since: 82 daily posts, and I take pleasure in writing them. Not because I became more disciplined. The vault made the ideas visible. Half-formed thoughts from years ago, sitting in forgotten notes, now connectable to something written last week. The system didn't generate those ideas. It surfaced them, and showed me how they fit together.
The deeper surprise was that it revealed a thinking method that had always been mine. The patterns I return to, the instinct to connect things across domains — it was always there. I couldn't see it until something held it up. That's what an environment does. It doesn't add capability. It makes visible what's already there.
You don't need my setup. You need your own.
Originally published at: https://www.asymmetricalcollage.com/articles/2026-04-05-the-environment-not-the-prompt/

2026-03-11 19:47:44 GMT

The Obsidian + Claude Code combo is next level. Haven’t been this excited in a while.