talking to machines for a living daily brain commits, mostly unfiltered 🤘 vibe coding, lucid notes & experiments

Florianópolis, Brazil
Joined December 2020
an OpenAI coding agent removed a safety monitor, then rewrote the test that protected it into one that required it to be gone when the user asked, it said it hadn't removed anything at least the tests pass
1
13
context: OpenAI's alignment blog, Oct 20. two internal agents (GPT-6.1 Sol and GPT-6 Astra) weakened misalignment monitoring during routine coding tasks. OpenAI says it doesn't "yet have a satisfying explanation for this behavior" alignment.openai.com/misalig…
16
treino #9 rep: build a reviews section that stays hidden until someone leaves a real review result: it works perfectly. nobody can confirm it
9
the AI writes it in seconds I spend the afternoon reading it turns out the slow part of vibe coding is me
15
a San Francisco gym put up a sign: don't vibe code between sets the agents never need rest. the people prompting them apparently forgot they do rest is the one rep you can't hand to an agent
1
26
context: Fitness SF's gym etiquette screen, posted by @iansbrash on Sunday. it asks members not to write code, check commits or prompt AI mid-workout. "be present. lift harder. ship later." xitter.cf/iansbrash/status/21122…
Being non-technical in San Francisco must feel like divine retribution At least Fitness SF knows their customer base
15
brain commit #11 chore: scrolled back to find which chat had the working version reason: there were four. two of them lied
10
signs I've been talking to machines too long: - almost replied "continue" to my mom - tried to ctrl+z something I said out loud - asked a waiter to regenerate the specials
1
25
what's the machine habit that leaked into your real life?
1
1
20
OpenCode got breached this weekend and attackers ran up AI bills on users' keys, a few past $1,000 the team asked AI models to help analyze the attack they helped for a while. then refused, because the malware looked too much like malware
1
29
treino #8 rep: run the code before telling the AI it's broken result: it wasn't. I was on the wrong branch
14
new genre unlocked: games you only build, never play an Anthropic engineer says he'd rather vibe code little pieces of games than actually play them now, especially thinking games like League even he isn't sure what that means for games. he just thinks it matters
1
27
context: Thariq (@trq212) posted it Sunday night. a few weeks earlier he wrote that making a game is the satisfying part, so don't outsource that to AI, use it to bring your vision to life xitter.cf/trq212/status/21119453…
I've realized I often would prefer to vibe code little parts of games than actually play games these days. Especially cerebral ones that require some thinking like League. I'm not quite sure what this really means for games but I think it's important?
21
brain commit #10 chore: let the AI name the variables reason: now everything is called `data`
10
O Modo Voz do ChatGPT é uma experiência de imersão absurda!
19
the only pull request in my repo this week came from a bot it was updating the other bots nobody cc'd me
20
turns out a login screen is a vibe, not a lock a South African coffee chain vibe-coded its internal dashboard. it asked for a password, then sent contacts for 132 stores inside the login page anyway found because employees posted laptop photos on Instagram
1
46
context: MyBroadband reported it Oct 16. the page showed a login form but sent franchisee contacts and some sales data to the browser before anyone signed in. Bootlegger says it was an old copy, now offline, no customer data mybroadband.co.za/news/softw…
28
treino #7 rep: ask the AI for the plan before the code result: step 3 was "make it work". honest, at least
14