Juraj

Every product owner or whoever manages feature descriptions in backlog should start vibe-coding. If you find yourself frustrated and struggling to describe the feature to the state-of-the-art coder model, please, stop for a while and think about how devs working with you must feel.

The LLM prompt is basically a set of constraints as in a constraint programming problem. The more details (constraints) are in your prompt, the less improvisation (and hallucination) will occur. This is direct analogy to solver landing in the local minima on the wrong side of search space.

“It works with my prompts” is the new “it works on my machine”. #llm #ai

Been tinkering with pipecat.ai lately. Such a well designed framework to work with. Trying to build multimodal context aware assistant one can self-host. 🤖 #pipecat #conversationalai #multimodal

Spent half day on inference script for image classifier. Is there a transformers-like pipeline for onnx models? I don’t want to install pytorch, so optimum won’t work here. #huggingface #onnx #pytorch

Having kids is obviously amazing. Don’t you love how your newborn wakes up the moment it touches the mattress after fifth round of you soothing it to sleep?

LLMs are actually stateless. You put in the historic context as input. Chat is only a UI constraint to build this context iteratively.

Today is my last day at company I’ve worked for two years. Learned a lot, met some great people. Now, I’m ready to start new year with new challenges and am generally excited about how 2025 will turn out. 🙂 Fingers crossed 🤞.

This will never get old 🤣

I’ve been exploring pocketbase.io for several weeks now and am quite impressed by it. Today I’ve extended it with custom JS proxy routes to “hide” other APIs behind its auth. What a nice tool to have in a tool belt! 🛠️ #webdev #api #backend

I'll get straight to the point. We trained 2 new models. Like BERT, but modern. ModernBERT. Not some hypey GenAI thing, but a proper workhorse model, for retrieval, classification, etc. Real practical stuff. It's much faster, more accurate, longer context, and more useful. 🧵

On Linux you're an admin On Mac you're an end user On Windows you're a victim

If you find yourself with too much free time over the (long) weekend / holidays, I have ~3h Building an LLM from the Ground Up workshop on YouTube that may come in handy: m.youtube.com/watch?v=quh7...

Anthropic released an interesting thing today: an attempt at a standard protocol for LLM tools to talk to services that provide tools and extra context to be used other the models modelcontextprotocol.io

I like this new analogy for working with LLMs by @emollick.bsky.social "treat AI like an infinitely patient new coworker who forgets everything you tell them each new conversation, one that comes highly recommended but whose actual abilities are not that clear" www.oneusefulthing.org/p/getting-st...

Few network resources disappeared from resource group on Azure. No logs or traces. Spent hour on call with support, no clue what happened. Now my VM is just sitting there, bricked, floating in digital space, waiting for rescue. ☁️ Just another Monday.

Major AI Large Language Models (LLMs) ranked by capabilities (MMLU) interactive https://informationisbeautiful.net/visualizations/the-rise-of-generative-ai-large-language-models-llms-like-chatgpt/

👋 Hello BlueSky! Another member just landed.