he Forest Can't Relax You If You Bring The City In Your Hand. You can be in the forest and get almost none of the benefit. I am talking about the phone in your hands. I do it constantly. Qing Li at Nippon Medical School measured it. ✅ A forest trip raised natural killer cell activity, your anti-virus and anti-cancer cells, by roughly 50%, and it stayed up for at least a week. ✅ Trees emit compounds called phytoncides, you inhale them, your immune system responds. ✅ A city trip did nothing. That part holds. What the viral version skips is the catch! ✅ Phytoncide uptake depends on your body being in parasympathetic mode, the calm rest-and-recover state. ✅ Walk through the trees head-down in your phone, and you keep yourself in the opposite state the entire time. ✅ The compounds are in the air. Your body just isn't set up to use them. Resharing because the reminder is aimed at me too.👇 https://x.com/Rencrypta/status/2073819567206375495?s=20
- 1 reply
- 1 recast
- 6 reactions
My Claude AI bot read 63x more than it spoke last month. 24.5M tokens in. 391k out. The bill was about $80, and the ratio is the entire lesson. I run an AI character that lives in a 3100 tg chat 24/7. After the June bill I did full cost surgery and wrote up every cut, with numbers: ✅ split the brain: Haiku for plumbing, Sonnet for personality only ✅ halve every context block (history 10 to 5, analytics 6000 to 4000 chars). Nobody noticed ✅ make replies roll dice on length: 55% one-liners. The character got MORE alive, not less ✅ video sight for pennies: 3 ffmpeg frames to the cheapest vision tier ✅ the hat-trick bug: two API endpoints describing one match disagree at the worst moment Full breakdown with the exact numbers and the bugs in the article below. 👇 https://x.com/Rencrypta/status/2072742579741855807?s=20
- 1 reply
- 1 recast
- 6 reactions
A Lot Of Hype About Loop Engineering Someone open-sourced a framework, Boris Cherny from Claude Code reposted it, everyone's bookmarking. I've had one running in production for weeks. My community bot The Undertaker doesn't just respond. It runs loops: ✅ twice a day it collects fresh intel on its own, match data, expert takes, form, and caches it ✅ it makes predictions, then grades itself against the real result and keeps a track record ✅ it feeds its own past hits and misses back into the next prediction, so it gets less wrong over time That last one is the whole point of loop engineering. A loop that doesn't check its own output against reality isn't a loop. It's a cron job with extra steps. The framework everyone's sharing is a great starting point. But the hard part was never the scheduler. It was teaching the thing to be honest about when it was wrong.
- 0 replies
- 1 recast
- 8 reactions