Elektrine lite

← Feed

@marius851000@mastodon.mariusdavid.fr

2026-02-17 08:05 UTC

@DoodleDonut@equestria.social Sad to hear this. Personally, I have absolutly no presure on my employer to use it (thought I still use it as it is usefull at times, I had to convince them that they stayed in full control of the code. I use an LLM running on my own computer. You can run fairly good model with an RTX 3090/4090/5090). If you are interested, here is a visuallisation of how they tend to technically work. (don't know if you are actually interested thought) https://bbycroft.net/llm

Replies (1)

  • @DoodleDonut@equestria.social Some more important technical stuff not shown in this visualisation is "reasoning"/"thinking", which allow the LLM have a sort of scratch space it can use to "think" ahead of time before generating the result, and tool calling, which allow the LLM to pause while providing some JSON that the interpreter interpret as the LLM wanting to call a function (it has been allowed to perform) and provide it the output (or an error). Its how they interact with their environment.

    Open ##2373320