Post · bonfire.cafe

RE: https://mastodon.social/@ctietze/116503658915086513

Nowadays, I almost never use a separate terminal as I cover all my use cases with vterm/eshell inside #emacs. While vterm is good, it has some quite rough edges here and there.

Recently, I was playing with #ghostty and it was super performant, standard complaint and just felt really nice. I guess ghostel can become my goto choice, when we properly package ghostty for #Guix.

I'm really glad to find this project, thanks to @daviwil for boosting the post 👇

Christian Tietze

@ctietze@mastodon.social · 2 days ago

I think @mitchellh and contributors' libghostty is an amazing gift.

I never had such a robust experience with Emacs terminal emulation like I have with https://github.com/dakra/ghostel

The rendering of things I use Just Works™️

#ghostty #guix #emacs

Debacle

@debacle@framapiaf.org · 3 hours ago

@abcdw @daviwil

Too bad, that #ghosttty will not be in #Debian any time soon, so neither will #ghostel 😞

Reason: We don't have #zig, and it doesn't look promising to me.

I'll stay with #eat and #vterm 🤷

#ziglang #Emacs

Chip Butty

@otfrom@functional.cafe · 4 hours ago

@abcdw @daviwil I'm probably not going to depend on something claude written if I can avoid it (and it is becoming more difficult to do so)

Andrew Tropin

@abcdw@fosstodon.org · 3 hours ago

@otfrom This statement and topic is quite tricky 😊 I would be glad to elaborate, but it will take a few pages.

In short, there are multiple valid concerns about LLMs in general and providers in particular, but telling "I avoid something just because some specific tool was used during the development" is obscure and unconvincing.

In such discussions, I suggest to specify exact reason(s), this way it won't sound like a blind belief, but a rational thoughtful decision.

@daviwil

Chip Butty

@otfrom@functional.cafe · 3 hours ago

@abcdw In this instance, I'm avoiding claude. The main reasons I try to avoid LLM output in code are:

Programming is Theory Building (see Naur 1985). If the LLM is doing it then no one has a theory about the code and there is no one to ask.
Copyright and licensing issues around LLM produced code. There are still some issues around whether or not LLM produced code is violating copyright (tho some settlements with non-code authors on this recently), so LLM code is a risk. Also the code itself might be in the public domain rather than under a Free license, but again it is a bit up in the air.
LLM scrapers are making it more difficult to run internet services due to aggressive scraping.
LLMs as composed by large companies are anti-worker in what they do and are damaging the livelihoods of those around me.
LLMs are causing a massive expansion of water usage and CO2 production in their training and output.

There are plenty others, but these 5 are sufficient. I'm not a purist here (I know that claude was used as I have it blocked on github where I do a lot of my work). But I'm trying to make better choices.