choas

Mein LinkedIn Learning Video ist online: IoT-Grundlagen für Entwickler:innen – Elektronik-Basiswissen linkedin.com/learning/iot-g… #Arduino #Iot #Elektronik

"AGI" -- written by a human

@ThePrimeagen 0 days since AGI took over the world

It's been 0 days since AGI has been achieved

@theo yes, build Dirigent (German for conductor) which executes prompts and don't have an IDE, but a diff viewer and "cue" tasks

Fwiw, I think everyone is going to do this now. It’s a much better UI for the “new way of coding”. Cursor’s main UI has been in a rough state for like a year now. I really like that they are resetting. Overdue tbh. Also really like that they support ACP, might have T3 Code support in near future? 👀👀 (Composer 2 is really good, don’t sleep on Cursor) (Also insert obligatory *I’m an early investor with early access* bias accountability)

Cursor just dropped a T3 Code clone

@asystemarchitec @paulo_kombucha yes, and I can toggle it

@choas @paulo_kombucha 😄OMG. That's awesome. A bit tiny, is there a way to scale it up?

Clippy is back. This time he's actually useful He watches your Claude Code agents, catches permissions, and jumps you to the right terminal

I just realized that Dirigent added also a a settings menu at the macOS menu bar

@asystemarchitec @paulo_kombucha I have now a Desktop lava lamp (like the pixel style) ... and it's controllable (start, stop, color, speed) via console, http, scheme

@infoxiao @kchonyc @karpathy I always tell Claude that "we" will fix this bug - waiting for the day when Claude tells me that this time it's my turn to do it.

@kchonyc @karpathy i once measured if using 'i' or 'you' made a difference for gemini. it did not. i was disappointed.

thanks to @karpathy , now i have cracked the mystery why my agent doesn't follow my instruction closely enough.

I'm cooking

@asystemarchitec @paulo_kombucha added one at the bottom right

@paulo_kombucha Well, I've been imagining a left-side panel, highly pixelated, like the claude logo when you start your session, and this pixelated lava lamp is doing its own thing like clippy, just a retro pixel thing. 😁

@paulo_kombucha wait, my Clippy uses a very small local open source model: a) fast b) stupid as Copilot c) helpful as Copilot maybe someone from #Microslop is interested?

@Haezurath tronos.dev - The AI-Native Operating System

Drop your project url Let's shout-out some future founders

@headinthebox wait, and they don't have a /flicker command to turn it on again?

Claude Code's flickering has disappeared, there are some weird artifacts in the input area where things get out of sync, but there is a new annoyance where in many case you cannot copy text from the output (for example when you see it go off the rails and want to past a snippet into the input field to steer it in the right direction). Also it often randomly jumps focus to the start of the chat.

@myfear because new Date(2026, 3, 13) is a Monday

Why do Java programmers sleep so peacefully on Friday the 13th?

@trashh_dev no, Elon thinks I'm a bot

are you an npc?

@fhinkel loop: output = call_model(prompt) result = validate(output) if result.valid: return output else: prompt = inject_error(prompt, result.error) continue

People complain that LLMs are unreliable. That’s a skill issue. You’re treating a probabilistic engine like a calculator and getting mad when it hallucinates. Vibe Coding expects perfection from a single call. Loop Engineering enforces it through recursion. You achieve determinism through verification loops. * Call the model. * Verify the schema. * If it fails, inject the error and retry. * Repeat until valid.

@theo waiting for T3 CLI

> “Does T3 Code support local models?” No. T3 Code is a serious developer tool. Locally runnable models are not capable of meaningful engineering work.

@ryancarson @varlockdev @syntaxfm added something similar to my Ralph CLI (podman / docker container gets an configurable .env)

Heard about @varlockdev on @syntaxfm and it looks good. Anyone using it? github.com/dmno-dev/varlo…