Off My Chest
RULES:
I am looking for mods!
1. The "good" part of our community means we are pro-empathy and anti-harassment. However, we don't intend to make this a "safe space" where everyone has to be a saint. Sh*t happens, and life is messy. That's why we get things off our chests.
2. Bigotry is not allowed. That includes racism, sexism, ableism, homophobia, transphobia, xenophobia, and religiophobia. (If you want to vent about religion, that's fine; but religion is not inherently evil.)
3. Frustrated, venting, or angry posts are still welcome.
4. Posts and comments that bait, threaten, or incite harassment are not allowed.
5. If anyone offers mental, medical, or professional advice here, please remember to take it with a grain of salt. Seek out real professionals if needed.
6. Please put NSFW behind NSFW tags.
view the rest of the comments
Yeah so I've not used claud but have used a number of models from hugging face.
I haven't used them extensively.
In my experience, they provide a great starting point for things I haven't interacted with much. So I might spend 10,000 hours with js, but never touched a firefox extension, or maybe a docker container, or nix script. With js an LLM is not much more productive than just coding by myself with non-AI tools. With the other things it can give you a really good leg up that saves a heap of effort in getting started.
What I have noticed though is that it's not very good at fine tuning things. Like your first prompt might do 80% of the job of creating a docker file for you. Refining your prompt might get you another 5% of the way, but the last 15% involves figuring out what it's doing and what the best way to do it might be.
With these sorts of tasks models really seem to suffer from not knowing what packages or conventions have been deprecated. This is really obvious with an immature ecosystem like nix.
IMO, LLMs are not completely without virtue, but knowing when and when not to use them is challenging.
This is where custom setups will start to shine.
https://github.com/upstash/context7 - Pull version specific package documentation.
https://github.com/utensils/mcp-nixos - Similar to above but for nix (including version specific queries) with more sources.
https://github.com/modelcontextprotocol/servers/tree/main/src/sequentialthinking - Break down problems into multiple steps instead of trying to solve it all at once. Helps isolate important information per step so "the bigger picture" of the entire prompt doesn't pollute the results. Sort of simulates reasoning. Instead of finding the best match for all keywords, it breaks the queries down to find the best matches per step and then assembles the final response.
https://github.com/CaviraOSS/OpenMemory - Long conversations tend to suffer as the working memory (context) fills up so it compresses and details are lost. With this (and many other similar tools) you can have it remember and recall things with or without a human in the loop to validate what's stored. Great for complex planning or recalling of details. I essentially have a loop setup with global instructions to periodically emit reinforced codified instructions to a file (e.g., AGENTS.md) with human review. Combined with sequential thinking it will identify contradictions and prompt me to resolve any ambiguity.
The quality of the output is like going from 80% to damn near 100% as your knowledge base grows from external memory and codified instructions in files. I'm still lazy sometimes and will use something like Kagi assistant for a quick question or web search, but they have a pretty good baseline setup with sequential thinking in their online tooling.