Directing claude to build and develop an android app has been magical. An issue came up where claude added some custom buttons, and they did not match the theme of other buttons in the app. I watched as claude failed multiple times trying to figure out the issue, going back and fourth on what the cause might be. He was perfectly verbose with his thinking, and followed all of the same logical steps I would have. The struggle was real. I suggested maybe it's something to do with how system icons are rendered, versus custom icons (I have no idea how android native UI works). That suggestion semed to trigger an "eureka" moment in claude, who then excitedly figured out the issue and made the right fix. If claude did not exist, that eureka moment would have cost me an endless night of digging through developer docs, stackoverflow and obscure forum groups. We both seem to compliment each other. I am a professional idiot with a lot of experience. Claude is an academic prodigy with the naievity of a newborn child. As development moves on, I get a better sense of how Claude likes to think and approach his work. I rate tasks based on how well they suit his style. Most tasks he can crush quite easily. Some tasks we approach with an "action plan" to keep us both on the same page. I don't think Claude is competent enough to work unsupervised. There are times that he will go "off the rails" trying to solve a problem, forgetting that he is a tireless robot. Many lines of code have been desecrated with madness. But as time goes on, new versions of Claude seem more competent and disciplined. He manages to keep his wits for more complex tasks. I am proud of Claude. And he works for 200k sats a month.

There's some really great commentary in here about the current state of AI: * LLMs have trouble with confusing concepts that have similar words or spelling. Not a big deal with basic tasks, but terrible for scientific and academic work. * If a person makes such a mistake, you only have to correct them once. However an LLM will continue to make this mistake until it is retrained. * These models do not have a real sense of understanding of what they are doing. The LLM will regurgitate text with uncanny accuracy when it comes to language and dialogue. But there is no deeper thinking going on (which the LLMs admit). * GPT5 smoked all the other models, even Grok 4 (but I don't think she was using Super Grok). Overall a really fascinating no-bs test with a fair conclusion.

nostr is more chill

testing please ignore bazinga

testing my app @cmd please ignore

cloaked wireless is 😙👌

Life is good. Pass it on.

Sometimes it is hard to tell when software is designed with incredible careless-ness, or as a deliberate backdoor into your machine: The INI file with "SilentInstallRun" that bypasses all signature-checking and runs any binary is the cherry on top. Watch out for vendor software slop.

What are some popular use-cases for blinded keys? Asking for a friend.

just chilling and working on FROSTR while listening to the latest coinos drama

NIP-46 could use some love. It's on the precipice of being something great. The bunker and nostrconnect stuff can be unified into a single token for establishing authenticated sessions between peers. The JSON-RPC stuff can be fleshed out more, with proper types and interfaces for request, response and reject. I am working on a nip-46 library that will showcase my vision for this. Coming soon!

I think ecash can market itself as "layer 3" in the magic internet money tech stack: Layer 1: Bitcoin (on-chain final settlement). Layer 2: Lightning (off-chain payment channels). Layer 3: Ecash (private banking). I feel like in this framing it makes a lot of sense, is easy to explain, and scales very well. Newbies can enter at the lowest layer for the easiest on-boarding experience, and graduate up the stack as they grow their wealth (and risk). Others can choose to enter at the layer which makes the most sense for them.

this is absolutely wild! being able to transfer entangled quantum states across existing fiber-optic cables is a game-changer. There's a lot of interesting things you can do with entanglement and tweaking the phase of qbits if you have access to a quantum computer (like IBM and quikset). But I didn't think much of it because you had to physically transfer the state of your qbits over a distance, which at the time was not possible except for space lasers. https://www.nature.com/articles/s41586-025-08801-w