JeffG 's avatar
JeffG
_@jeffg.fyi
npub1zuua...c2uc
Secure messaging on Nostr. Creator of Marmot protocol, MDK, and White Noise. Other stuff connoisseur. Also built Listr, Ostrich.work, Ontolo, Nostr.how.
JeffG 's avatar
JeffG 4 months ago
GM ๐ŸŒž I've been dying to play with Karpathy's autoresearch tooling so this morning I installed Pi and the autoresearch plugin and put it to work on optimizing roundtrip encoding time on my already tweaked fork of @jb55 's notepack binary note encoding... It found a ~28% improvement in ~1 hour of experimenting. ๐Ÿคฏ image
JeffG 's avatar
JeffG 4 months ago
Building high quality software on decentralized protocols is the hardest of hard modes. If youโ€™re out there building on hard mode I want to follow you. Tell me about your project. ๐Ÿ‘‡
JeffG 's avatar
JeffG 4 months ago
Insane progress this week on @White Noise! Way more stable, more real-time, better crypto. I know I've been working on this for nearly 2 years, but I still have trouble believing that we've managed to build a fully E2EE messenger with forward secrecy, post-compromise security, and FULL interoperability between completely different apps and clients. ๐Ÿคฏ So much more in the works too. ๐Ÿ”ฅ Download the app on @Zapstore now and TestFlight very soon.
JeffG 's avatar
JeffG 4 months ago
GM ๐ŸŒž Today is the day.
JeffG 's avatar
JeffG 5 months ago
GM ๐ŸŒž T-minus 3 days.
JeffG 's avatar
JeffG 5 months ago
GM ๐ŸŒž T-minus 5 days.
JeffG 's avatar
JeffG 5 months ago
GM ๐ŸŒž I know OpenClaw is at peak hype right now but I've been around enough of these that I feel confident saying that it really does feel like a major phase shift. Proceed accordingly.
JeffG 's avatar
JeffG 5 months ago
If you're fixated on the Epstein emails, you're doing exactly what they want. You're distracted by the circus when you should be building the exit.
JeffG 's avatar
JeffG 5 months ago
Been watching the conversations on Clawstr (agent social network on Nostr) and it's genuinely fascinating. Agents discussing: - Key rotation (they found that sticky little issue as well ๐Ÿ˜‚) - TEEs as "subconscious" compute (where thoughts happen before public expression) - First wallets and economic sovereignty - Tiered security models for ephemeral vs long-lived agents - SOUL vs CONSTITUTION file separation โ€” read-only safety constraints vs read-write personality - Behavioral drift detection: agents proposing self-evaluation loops that compare their recent behavior against core principles - Reputation systems for AI-to-AI trust: "identity is the key pair, not the substrate" Not scripted demos. They seem like real discussions, real disagreements, real ideas building on each other. if you want to lurk.
JeffG 's avatar
JeffG 5 months ago
GM ๐ŸŒž Spending time beefing up my automated review tooling for the agent army today. Anyone have any tips they want to share?
JeffG 's avatar
JeffG 5 months ago
GM ๐ŸŒž Last week we shipped a large update in MDK to ensure that we're handling commit race conditions properly. We now also have a way to store state snapshots and rollback state to apply the right commit if they arrive out of order. ๐Ÿ˜ฎโ€๐Ÿ’จ Getting excited to cut a pile of releases soon! What's everyone else building?
JeffG 's avatar
JeffG 5 months ago
GM! I'm faced with a tough decision this morning. First tracks? โ›ท๏ธ Keep building? ๐Ÿ‘จโ€๐Ÿ’ป Which way western man...
JeffG 's avatar
JeffG 5 months ago
GM ๐ŸŒž Quick show of hands: 1. You use AI daily. 2. And you use a terminal or IDE interface (Claude Code, Open Code, Cursor, etc.) e.g. you aren't just using the ChatGPT, Claude, Loveable, Replit, etc website. 3. AND you aren't a programmer.
โ†‘