Just canceled my @cursor_ai subscription after a year of usage.
It’s clear that if you don’t fully own your software and AI stack you will be subject to companies disturbing your workflow
Owning the hardware is great!
If anyone is interested in an homelab guide for your next AI project, I wrote a post a few months ago (featuring proxmox, gpu passthrough and initial software setup)
I’ve been doing it since the beginning all I need for any repo are
- a file tree
- a comprehensive text + diagrams readme
- a contributing file
- a changelog file
And finally the issue / feature to build
Each commit is a changelog block
ggerganov has it rough.
imagine writing the backend for your llm inference in C++...
then some rando group wraps it with a non-compatible API, calls it "ollama", which gets a majority of the outside attention and support
In my AI team we released today our first batch of agents using @AgnoAgi! (cc @ashpreetbedi)
The framework is super well thought out and integrates very well with our @vllm_project cluster
> Everyone uses CF
> CF disallows other AIs
> ID for agents
> Govs are pushing for IDs
There is almost zero path to decentralization
tldr: imo this isn't helping
I created a script that automates the setup of a fresh Ubuntu 24.04 server for AI/ML development work. It handles the complete installation and configuration of Docker, ZSH, Python (via pyenv), Node (via n), NVIDIA drivers and the NVIDIA Container Toolkit, basically everything
Running models on my own GPU, looking to buy more and on the software side leaning more into vscode / zed + custom ai cli with tools that I made a while back (like Claude code)
Building / Fixing according to architecture designs, requirements and issues
