Advertisement · 728 × 90

Posts by

Running OpenCode with my local Qwen3.6 35B A3B UD Q6 K setup and it’s way better than I expected.

Fast, holds context, and actually useful for real work instead of just demos.

Local models are getting very real. Curious what people here are running.

#LocalLLM #Qwen #LLM #AI #OpenCode

23 minutes ago 0 0 0 0

Well, it finally happened. We’re redesigning the website… for LLMs, not humans.

#AI #LLMs #SEO #FutureOfWeb

6 days ago 0 0 0 0
Post image Post image Post image

Went from a text prompt to a 3D model using ImageZ Turbo + Hunyuan3D.

Entire pipeline ran locally in ComfyUI and finished in under an hour.

Still rough in spots, but the pace of this is hard to ignore.

#ComfyUI #LocalAI #AI3D #GenAI

1 week ago 1 1 0 0

It would be nice to go a full week without AI-written code breaking production.

We *had* 7 years without any prod issues before the mandate to increase velocity by replacing people with AI.

#AI #DevLife #SoftwareEngineering #TechReality

2 weeks ago 1 0 0 0

My OpenCode suddenly stopped working with Claude.

It’s so much nicer than Claude Code that I don’t want to go back.

Tried generating a new key a few times, just keeps failing. I’m on a paid account. I asked Claude about it and it took me to a page asking for $5

#AI #LLM #DevLife #Claude #OpenCode

1 month ago 1 0 1 0

We’re in full “AI must write the code” mode at work.

The system has never been more unstable.

Every morning starts with a long list of bug reports and “run this in the terminal” fixes.

A junior would learn from this.

Management won’t.
AI or bust.

#AI #LLM #DevLife #AIReality #Tech

1 month ago 1 0 0 0
Preview
GitHub - ripter/luaLLM: llama.cpp runner to make running with a config easier llama.cpp runner to make running with a config easier - ripter/luaLLM

Yeah, still very much WIP.
github.com/ripter/luaLLM

Running models (single or server) works great and I use it every day.
You can keep notes on models, and status --json makes it easy to script around.

Auto profiling doesn’t quite work yet.

1 month ago 1 0 1 0

Been using Qwen3-Coder-Next-Q8 as my main local LLM. My CLI tool makes it trivial to try other models which is nice. Powered by llama.cpp

1 month ago 2 0 1 0
Advertisement

I built a CLI to make it easier to run local LLMs with llama.cpp.

A few weeks later I have another tool using it to read my email and apply labels with a local model.

No credits, just scripts and a model on my laptop.

This is what I wanted local LLMs for.

#llm #localllm #llamacpp #devtools

1 month ago 7 0 0 1

Been playing with OpenCode + Claude and I’m starting to understand the people who say programming is dead.

I don’t agree…

but the improvement over the last 4 years is honestly a little scary.

#AI #LLM #Coding #DevLife #AIReality

1 month ago 3 0 1 0

Work is switching to Copilot.
Probably because of contracts.

Nobody cares as long as the LLM is still Claude.

We’ve reached the point where the model matters more than the IDE.
Might as well go back to vim. Hello again old friend.

#AI #LLM #DevLife #Copilot #Claude #Vim #AIReality

1 month ago 1 0 0 0

Claude writes a lot of code.
Also breaks a lot of code.

The best feature is when it decides I’m not allowed to code anymore today.

Out of credits until 5pm.

Update: 10pm.

Cool. The tests are still broken, but at least the AI is well rested.

#AI #LLM #DevLife #AIReality #HumanInTheLoop

1 month ago 2 0 0 0

LLM coding is great for those little quality-of-life updates you’ve always wanted to do but never make time for.

Like improving that script where you always forget the arguments, so you add a picker and some basic sanity checks.

#AI #LLM #Coding #DevLife

1 month ago 1 0 0 0

Oh I did that Mr AI. Several times, even tried different models.

1 month ago 0 0 0 0

Two days debugging a test with Claude.
Code rewrites. Debug prints. Same “fixes” on repeat. Credits gone.

I finally looked at it myself. Took me less than an hour to find the bug.

Bad import statement.

#AI #LLM #Coding #DevLife #AIReality #HumanRequired #Debugging

1 month ago 1 0 2 0

Stirling engine → 10Ω → Schottky diode → 220F supercap.

Ran for 24:42 on 8g of Denatured Alcohol.

Cap went from 1.19V to 1.2885V.

Stored ~27 joules. ~18mW average.

#electronics #StirlingEngine #DIY #EnergyHarvesting #Stirling #DIYenergy #supercapacitor

1 month ago 2 0 0 0

Nothing like a 2026 AI confidently gaslighting me about a 1994 SNES game.

ChatGPT: “You don’t need to go back to Threed.”
Me: stuck in Mr. Saturn Village for days.

An ancient walkthrough: “Go back to Threed.”
It worked.

#EarthBound #SNES #RetroGaming #AI #LLMs #GamingLife

1 month ago 2 0 0 0

Just hit the part in The Wandering Inn: Lady of Fire where Magnolia enforces rules on Erin’s door without consent.

Grrr, if I were Erin I would cut off Invrisil. The door can move to another city. But I’m not a Chessmaster like Erin.

#TheWanderingInn #LadyOfFire #MagnoliaReinhart #ErinSolstice

1 month ago 0 0 0 0
Advertisement

Worst case, I containerize the house and ship it.

1 month ago 1 0 0 0

Generated code runs in sandboxed libs with strict allow/block rules for file + network access. This catches things my human brain might miss and prevents hallucinations from doing damage.

It generates tests and code. I hand review both. I think of it as Human Required.

1 month ago 0 0 0 0

Been working on my own agent program.
It can now generate its own code and use existing project files as context.

All running locally. Real local.
No API keys. No third-party bills.

#AI #LocalLLM #Agents #Programming #IndieDev #SelfHosted #BuildInPublic

1 month ago 11 1 2 0

That feeling when you’re reading a good book and the final battle has started, but there’s still a third of the book left.

#Books #Reading #Fantasy #SciFi #ExpeditionaryForce

2 months ago 1 0 1 0
Post image Post image

Ordered 1800flowers for Valentines Day like I have many times before. And this year the flowers are very disappointing. Wife even said that grocery store flowers would have been better.

#VDay #Valentines #Flowers #Disappointing

2 months ago 2 0 0 0
Video

Here it is solving the "maze".

2 months ago 0 0 0 0
Post image

Been working on some reinforcement learning.
Got the model solving my basic maze.

Started with Claude, burned through days of credits, still couldn’t get it there. Ended up writing it by hand.

Now it works. 8 headless Godot servers running physics in parallel.

#RL #AI #Godot #Programming

2 months ago 3 0 1 0
Advertisement

AI has 3 phases:

1. “It’s amazing!”

2. “Dumb POS” (you waste days on what you could’ve done in hours)

3. Acceptance: useful sometimes, nowhere near the hype.

#AI #LLM #DevLife

2 months ago 0 0 0 0
Post image

After 4 hours of running bright, I’ve gotta call it and go to bed.

Cap is still sitting at 1.88V, which means in theory this LED would’ve stayed bright for another hour+ easy.

#electronics #supercapacitors #ultracapacitor
#benchlife #diyelectronics #learningbydoing
#energyharvesting

2 months ago 0 0 0 0
Post image

More than 10 min since I disconnected the batteries and it’s still going strong.

#electronics #supercapacitors #ultracapacitor
#diyelectronics #benchlife #madscience

2 months ago 0 0 0 0
Post image

Let there be light! Got the cap up to 1.6V after many many hours.

2 months ago 0 0 0 0
Post image

Slowly, carefully, responsibly charging an ultracap for the first time. 100R resistor and 2xAA batteries.

Current-limited. Meter watching. No sparks.

#electronics #supercapacitors #ultracapacitor
#diyelectronics #energyharvesting #benchlife
#madscience #candlepower #learningbydoing

2 months ago 0 0 0 0