My first real hardware project. I've built my own RGB controller using @arduino. The color changes depending on hardware utilization (blue = RAM, green = CPU, red = GPU).

I'll publish a blog post with details soon.

In the meantime, here is what happens when I chat with Ollama.

#hardware #arduino #iot #tinkering

A second small update to Embeddings Playground. Now you can upload simple text files with one text input per line. Check it out here: https://embeddings.svana.name/

#ai #embeddings #machinelearning #ml #llm #nlp

A small update to my Embeddings Playground. I added 4 more original sentence-transformers models. Check them out: https://embeddings.svana.name/

#ai #embedding #machinelearning #llm

Do LLMs hallucinate more in Czech than in English? I ran some benchmarks, and it looks like the answer is yes: https://svana.name/2026/02/do-llms-hallucinate-more-in-czech-than-in-english/

#ai #aisafety #llms #ml

Here’s FileChat in action — a local, read-only AI coding assistant. Chat with an LLM about your codebase, without risking unwanted changes.

Alpha release → feedback welcome 🙏
🔗 https://lnkd.in/eBBt9f9X

#AI #DevTools #LLM

I’m working on my own opinionated local AI coding assistant: FileChat.

It doesn’t write code, it only reads it — exactly how I like it. I want to talk about my projects with an LLM without letting it touch a single line.

It combines a local embedding model running with chat powered by Mistral AI.

Still early days, but here’s a short demo.

#AI #LLM #DevTools #Coding

"Concrete Problems in AI Safety" is one of the most famous scientific papers on AI safety. But is it still relevant, given that it was published in 2016?

Check out my first post from the "Reading club" series to find out: https://svana.name/2025/07/reading-club-concrete-problems-in-ai-safety/

#ai #aisafety #artificialintelligence #llm #openai

I just successfully implemented my first reinforcement learning experiment, and I felt like a little kid.

The task was adapted from Sebastian Raschka, PhD's book, "Machine Learning with PyTorch and Scikit-Learn.". To win, an agent represented by a blue square has to get to the reward represented by the golden triangle. But if it steps on a trap represented by a black circle, it loses the round immediately.

#ai #ml #learning #reinforcementlearning

Another update to my Embeddings playground. You can now select between PCA and tSNE for embedding projection. And I am proud to say, that I implemented tSNE from scratch myself.

Also implemented a few small fixes related to model and reference text selection.

Check it out here: https://embeddings.svana.name

#ai #llm #ml #programming