diff options
Diffstat (limited to '2026/talks')
| -rw-r--r-- | 2026/talks/llms.md | 68 |
1 files changed, 46 insertions, 22 deletions
diff --git a/2026/talks/llms.md b/2026/talks/llms.md index d9256426..5a2941e1 100644 --- a/2026/talks/llms.md +++ b/2026/talks/llms.md @@ -19,6 +19,10 @@ Participants: [[!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 @@ -43,35 +47,55 @@ 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? What - freedoms it would need to have? Feel free to answer anything, not - limiting yourself to current LLM architectures. If your answer is - that no such software should exist, that is a valid answer. - -2. What is the most useful application of LLMs with Emacs for you - personally? If you are not using LLMs, imagine that there is an - ideal AI or complex automation software (non-AI) you would be - willing to use and come up with such application. - -3. What is the most critical use case in your Emacs workflows where - you would **avoid** using AI? - -4. Imagine that there is an LLM-generated code submitted to Emacs that - implements long-awaited feature (for example, multi-threading - support in Emacs or time zone support in Org). As a maintainer, - under what conditions, if any, would you accept LLM-generated code, - and what would need to change for you to accept it? As a user, - would you be willing to use such a new feature? You can answer as - a maintainer, user, or both. + 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 (for newcomers or otherwise) around LLMs/AI in the next - year? And one thing it should decline to do? + 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** |