summaryrefslogtreecommitdiffstats
path: root/2026/talks/llms.md
blob: 5a2941e1454b92dfb204437097fa0922eac8e7bb (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
[[!meta title="Panel: Emacs, LLMs, and Free Software"]]
[[!meta copyright="Copyright © 2026 Participants"]]
[[!inline pages="internal(2026/info/llms-nav)" raw="yes"]]

<!-- Initially generated with emacsconf-publish-talk-page and then left alone for manual editing -->
<!-- You can manually edit this file to update the abstract, add links, etc. --->


# [Panel] Emacs, LLMs, and Free Software

Participants:

- Alvaro Ramirez [known by agent-shell.el]
- Andrew Hyatt [known by llm.el]
- Divya Ranjan [known by canvas patch and Human Emacs]
- Karthik Chikmagalur [known by gptel.el]
- Kent M Pitman [known as early Emacs contributor, MACLISP, and Common Lisp developer]
- Screwlisp [known by lisperGopherClimate show]

[[!inline pages="internal(2026/info/llms-before)" raw="yes"]]

 **Important**: The views expressed by the panelists do not
 represent official position of FSF and the GNU project. They are no
 less and no more than individual views of the invited panelists.

The topic of AI (as a broad concept) and its flagship technology -
LLMs - is rather controversial. Some of us enthusiastically accept the
technology and its potential to change lives for the better; some are
more cautious, or outright question the long-term consequences for
individuals, countries, and the whole world; raise ethical concerns,
or point to high energy consumption (especially concentrated in USA).
However, there is no doubt that Emacs (and the whole world) cannot
ignore the topic.

The Emacs community already has both proponents and opponents of
the LLMs and a number of LLM integrations have appeared over the
last few years.  Here, we have brought a number of Emacs users
with very different views about AI together, to make the range of
positions explicit and to explore how Emacs and its community can
support different approaches without fracturing.

We tried our best to lay out the questions constructively, and to
avoid the general topics that have already been beaten to death
in online discussions across the web. We ask both the panel
participants and the EmacsConf audience to approach the
discussion the same way: with the goal of understanding each
other's positions and finding ways for different choices to
coexist.

Keep in mind that "AI" and "LLM" terms can often mean a number of very
different things. For example, LLMs can mean large language
model-based text generators, text embedding search models, voice
recognition, or even anything related to the field of neural networks
or machine learning. There is also often a confusion with people using
the word "LLM" referring to the most popular proprietary generative
models, although there are also open-weight language models or
language models that can run locally.  "AI" is also often used
interchangeably with "LLM" while it can also refer, say, to symbolic
artificial intelligence systems.

The questions below will use "AI" as a broad term that covers
everything, not just the popular large language models.
The "LLM" models will include large language models that can generate
text, not limited to the popular SaaSS proprietary models, but also
including local LLMs and open-weight models.

P.S. There is no system but GNU, and Linux is one of its kernels :)

Questions:

1. If you were to imagine an ideal AI software you would use with
	Emacs, how would it look like? How would it **not** look like?
	Would it be LLM text generator or entirely different architecture?
	What freedoms would it need: free/libre source code, open weights
	and training data if it is an LLM model, or something else?
	If your answer is that no such software should exist, that is a
	valid answer.

2. There is a broad range of uses of AI within Emacs. Pick an
	Emacs-related scenario that you find especially exciting or
	worrisome and offer us a sense of why that is capturing your
	attention.

3. LLM-generated text can show up in many places not limited to
    source code (commit messages, documentation, mailing lists,
    etc). What do you think about harms or benefits (if any) of
    LLM-generated text in non-code scenarios?

4. Imagine a partly or fully LLM-generated code submitted to Emacs
	that implements a long-awaited feature, for example
	multi-threading support in Emacs or time-zone support in
	Org. How would you deal with such submission as a maintainer or
	as a user? What motivates your judgement? You can answer as a
	maintainer, as a user, or as both.

5. What is one thing the Emacs community should build, document, or
	support around AI (be it LLM chatbots or other machine learning
	tech) in the next year? And one thing it should decline to do?

**Note that you can ask other questions not limited to the above
during public Q&A session after the panel discussion**

Related:
<https://lists.gnu.org/archive/html/emacsconf-org/2026-08/msg00007.html>


[[!inline pages="internal(2026/info/llms-after)" raw="yes"]]

[[!inline pages="internal(2026/info/llms-nav)" raw="yes"]]