predkambrij

Small fun toy. Infinite inspiring quotes generator. https://github.com/predkambrij/inspire

Yeah. The purpose of tests is to verify correctness of their setUp mocks :)

I have chromebox (with 32GB DDR4) that idle at 4W, but after adding couple nvme drives it doubled it's power consumption. Having full ATX mobo is cool (flexibility), with BIOS settings, powertop, and some other settings can also idle at quite low power. I have i7-7700K that idle at 18W. With combination of wake-on-lan and similar you can have a monster but won't empty your wallet.

/etc/anacrontab can solve your use case.

One cool trick is having (public) subdomains pointing to the tailscale IP.

Please, please. I'd love to use it with Debian.

This is exactly what was in movie Common People (Black Mirror). Every response included AD.

My experience with Perl. "Write-only" language.

I "skipped" couple chapters :)

I must admit that I didn't read the article in full.

I share your view. .keep and .gitignore are different things. Having one .gitignore caputuring everything is less mental load.

Messenger.com helped me to break from FB addiction. Now, I assume they want to convert some people to be FB users. I learned my lession.

Cool https://m.youtube.com/watch?v=hMS30w23zkQ

17t/s on a laptop with 6GB VRAM and DDR5 system memory. Maximum of 100k context window (then it saturates VRAM). Quite amazing, but tbh I'll still use inference providers, because it's too slow and it's my only machine with "good" specs :)

    cat docker-compose.yml
    services:
      llamacpp:
        volumes:
          - llamacpp:/root
        container_name: llamacpp
        restart: unless-stopped
        image: ghcr.io/ggml-org/llama.cpp:server-cuda
        network_mode: host
        command: |
          -hf unsloth/Qwen3-Coder-Next-GGUF:Q4_K_XL --jinja --cpu-moe --n-gpu-layers 999 --ctx-size 102400 --temp 1.0 --top-p 0.95 --min-p 0.01 --top-k 40 --fit on
    # unsloth/gpt-oss-120b-GGUF:Q2_K
        deploy:
          resources:
            reservations:
              devices:
                - driver: nvidia
                  count: all
                  capabilities: [gpu]

    volumes:
       llamacpp:

I'm a Linux user (I don't know Apple much). I use devcontainers (the repo I listed) for almost everything, because almost everything is possible to do with about zero overhead. Eg, it won't do memory fragmantation that VMs do, processes are just the same, as on host. GUI apps work just the same as the ones on host, which is really cool. Storage can be an issue if images multiply too much, but not a real issue. Incus is a bit different, I still evaluate it, so can't comment much. By description is just the very thing I was looking for (a VM, but without VM's overhead). It's amazing, that Debian12 desktop will start in about 2 seconds. Performance is again just the same as on host's machine. VMs are still cool, I still use them, where I want to be really sure that I won't have a bug that containers would cause them (weird networking and such).

I'm playing with repeatable development environment with incus. So, it's like in Docker, but naturally more things should be possible (eg. snap package manager is the ultimate test), but still more disposable than VMs (eg. it won't finish your laptop battery and annoy your ears with fan).

https://github.com/predkambrij/incus-container-desktop https://github.com/predkambrij/devcontainer

"Hold for Me": uses LLM to detect when somebody picks up your call https://github.com/predkambrij/Hold-for-Me-using-ADB

Speed: 217ms 289ms

I have 5G network :)

Also, heard multiple times that edge network can be worse, because if you're low prio and other part of globe is not busy, you get it routed in worst possible way.

Things about those approaches did and will change more when LLMs are getting better. I got some unbelievable good results back in March, then I was tasking LLMs too hard problems and got bunch of frustrations, then learned to prompt better (to give LLMs methods to test their own work). It's an art to do good balance of spending time writing prompts that will work. A prompt could be "fix all the issues on Github", but maybe it going to fail :)

Devcontainer: makes it easy to start working on random projects without installing a bunch of dependencies on the host computer.

https://github.com/predkambrij/devcontainer

Personal productivity app. Mirrors contents of a file (TODO list and other info) on your phone. You can make a call with Grok to brainstorm specifics. The goal was to remove any friction possible, to make focus easier. https://github.com/predkambrij/focusapp

I recommend "YouTube Summary with ChatGPT & Claude" extension, which makes it easy to copy transcription. Paste it in Deepseek or your favourite LLM with prompt "summarize in 1 paragraph" and then you can run TTS and you get idea about of the video in 10 seconds. You can ask followup questions, or to format it differently or get the reason to actually watch it in full.

CSV is a lot lighter on tokens, compared to json, so it can go further, before a LLM gets exhausted.

Another huge thing is that if you get banned from Google you (might) also lose "Sign in with <bigcorp>" - so you lose access to a lot more thing than just your email.

Sounds really interesting, very much new way of thinking. AI can ask questions to resolve ambiguities (sometimes might not matter - prototypes, and let it make best guess). Sounds a little bit like declarative programming. Spec is in a way also a test (is the button really red if it was specified red. What is red (should it be really red, or just a little bit). A lot of interesting challenges, but sounds really what could be helpful in the future.

you can set search?q=%s&udm=14& for google and default to "web" tab, where you actually get search results. Works fine for me.

I like it. Thank you for showing.

I regularly create a new folder with current date and work in it until it gets overbloated and then create a new one. Past ones remain until they are still relevant. If something is generally (evergreen) relevant, it gets placed in a text "longnotes.md" where they can also be searched and also found roughly by its date. Do whatever works for you.

seems to be straight "yes" with no opt-out. https://github.com/google-gemini/gemini-cli/blob/main/docs/t...

Interesting, it works for me through openrouter, without configured openai integration. Although, I have openai account and did verification with openai before. Conspiricy theory would say that they are exchanging PII so openrouter knows who am I :)