cortesi

cortesi

@cortesi.bsky.social

I make software, break software and make software that breaks software. https://corte.si https://mitmproxy.org https://binvis.io

11 followers13 following16 posts

A real conversation with my wife. If you marry right, you'll never be bored, fellas. Me: What are you up to? Her: Making decorations. What do you think? ☺️ Me: Oh my god Her: Do you still love me and appreciate my craft?

We just discovered this great lump of fluff, curled up peacefully as if sleeping in our garden, dead. He was the Best Cat, and we're overwhelmed with grief. Farewell, Kaiju.

Gaze upon this monstrous 14kg Christmas ham, fellow mortals. It's been slow roasted for 18 hours while being lovingly slathered with pineapple and maple glaze every 30 minutes, and it's headed straight for your midsection. All shall love it and despair.

Pro tip: work on UX when you haven't slept properly. Got 3 hours last night and the hit to my IQ means I'm now baffled by the complexity of my own tooling. Time to simplify so I can understand it even in my degraded state.

This is an incredible model - in my personal benchmark it's well above anything from OpenAI, and through a host like DeepInfra it's almost too cheap to meter for my personal use case. And it's licensed under Apache 2! Amazing.

Lukasz Olejnik@lukaszolejnik.bsky.social

With Qwen-2.5-Coder 32B LLM model, you can now have GPT4-grade model on a laptop (well, assuming that you have lots of RAM), which is sufficient for coding tasks. That's the best privacy-preserving LLM model out there: running on the device. huggingface.co/Qwen/Qwen2.5... arxiv.org/pdf/2409.12186

You are up to date.