christian faubel

Writing these lines while experimenting with Chatbots within Emacs, using both gptel and aidermacs. I am still trying to find an attitude towards the use of tools and possibilities provided by large language models. My own parameters or constraints on working with LLM are based on being curious and being autonomous. So I am happy i have access to some models hosted on German servers within my academic networks. While this is not self-hosting in the strict sense, it still gives me a large sense of agency, because at least in theory i could host these open-weight-models myself.

Having access to a chatbot directly within Emacs through gptel empowers me and allows me to get quick responses on asking about short-cuts and commands i don’t remember and it feels like it is supporting my learning journey within Emacs.

As a teacher, being surrounded by students who use the commercially available solutions from OpenAI, Google or Anthropic, i feel very much a responsibility to explore alternative routes.

Still within my Fediverse-bubble, I read a lot of strong rejection of any usage of LLM’s so writing these lines feels like a confession.

However my experience so far is rather positive, Emacs is becoming itself much more usable and i am starting to play with aidermacs for simple playful projects like hacking a defunct laser-cutter and turning it into a drawing machine.

# #

May 30, 2026

Comments

@c Since self-hosting or locally running so-called ‘open-source’ models only solves a minority of the many issues with LLMs, and in turn LLMs do not actually solve any real problems in practice, simply avoiding them entirely seems like the preferable option.

I guess I should have been more explicit on the constraint of curiosity, I am well aware of all issues around the training of LLMs, and I surely don’t want to appear as an LLM-evangelist. I understand that the notion of what can be considered a real world problem is broad. For some people, surely, turning a defunct laser-cutter into a drawing machine is not a real problem. But curiosity as a constraint also on the discussion of the issue at stake, would rather afford asking questions versus making statements.
I am well aware of the attitude that LLMs as a technology should be avoided at all, but here I do not agree, I think even for my own understanding why there is such a pull for using these technologies, it is mandatory for me to try them out. In addition, my curiosity comes into play the more something seems to be prohibited the more I want to play with it.

Reply

@c I don’t think it’s true that it’s necessary to use a thing in order to understand it; or more generally, to have personal experience of it.

There are many things which I don’t have personal experience of, which I can understand sufficiently to say why they’re good or bad, and why they’re popular.

Yes you are right, when I mentioned my understanding I was really meaning my own understanding and in my personal experience I get grasp on things by trying them out. This was not meant to be universal truth but a personal one. This is one reason why I like computers because they allow me play, to play things out. Like for example watching a self-organizing map restructure itself to find the Hue as a lower dimension of the RGB colorspace, or to watch a Perceptron learn Digits. The root for wanting to play is #curiosity and for me #curiosity is my daily driver.

Reply

@c

I can understand being a teacher why you need to at least be familiar with LLMs. but you need to know the cons as well as the pro's & articulate them with your students, the use by large us companies to manipulate the politics of European countries, the vast environmental impact of the data centres, digital sovereignty impacts & the biggest issue that everyone has under the input box for their AI the warning that the answers are not always right.

1/2

@c

LLms should be useful for local stored models & the creation of documentation, but anything generated by LLMs has to be checked by the person generating the output.

@dekkzz78 @c I would rather have no documentation instead.