Ihor Radchenko (yantar92)

Maintaining Org mode (GNU Emacs)

Public Key

npub1d52pjts7s5ypf2n955pjf8djns7eh29v75vfg8dnhx433t44yzxsf5eu3h

NIP-05 Address

Profile Code

nprofile1qqsx69qe9c0g2zq54fj62qeynkefc0vm4zk02xy5rkemn2cc466jprgpz3mhxue69uhhyetvv9ujuerpd46hxtnfduqs6amnwvaz7tmwdaejumr0dsu6fptt

Publishing to

relay.damus.io

nos.lol

Author Public Key

npub1d52pjts7s5ypf2n955pjf8djns7eh29v75vfg8dnhx433t44yzxsf5eu3h

Show more details

Last Notes

2026-11-01 18:26:32 GMT

This month in Org - November 2026 edition.
https://blog.tecosaur.net/tmio/2026-11-01-previewing.html
For code enjoyers: https://karthinks.com/software/understanding-org-latex-preview/
Original announcement by Timothy: https://list.orgmode.org/[email protected]/T/#u
#orgmode #latex

2026-10-17 10:00:31 GMT

#orgmode LaTeX preview landed on main. Nuff said.
Thanks @nprofile…24wu, Timothy, and all the people helping to test the previews for their efforts.

2026-10-17 07:41:45 GMT

@orgmode has a new review policy for complex patches.
From now on, complex patches a approached differently from other. They can be merged after fixing the most general issues without trying to fix smaller details. Smaller details are left to after the merge.
Announcement: https://list.orgmode.org/87ik30u72v.fsf@localhost/T/#u

2026-10-12 18:40:25 GMT

From now on, Org mode depends on compat.el [1] to handle backwards-compatibility.
Thanks to Morgan Smith who finalized the patches, and to everyone participated [2], including @nprofile…muy0, the author of compat.el.
(It took the casual 5 years to complete).
#orgmode #emacs
[1] https://elpa.gnu.org/packages/compat.html
[2] https://list.orgmode.org/orgmode/[email protected]/T/#u

2026-10-12 16:25:52 GMT

@nprofile…wsge I hope I got your email right ;)

2026-10-11 14:15:12 GMT

Going to experiment with panel discussions during the upcoming EmacsConf:
1. Emacs, LLMs, and Free Software: Access, Agency, and Accountability
2. Improving the Emacs Community: Onboarding, Participation, and Growth
Some speaker invitations already sent :)
#emacsconf
CC: @nprofile…9y77

2026-10-10 14:03:12 GMT

#5 OrgDevMeetup on Sat, October 17, 19:00 UTC+3
Org *development* meetup is scheduled next week.
The time should be suitable for US.
The meetup focuses on actual Org development, live, together.
More information: https://orgmode.org/worg/orgdevmeetup.html
URL: https://bbb.emacsverse.org/rooms/orgdevmeetup/join
#orgmeetup #orgmode #emacs #meetup
CC: @nprofile…9y77

2026-10-09 17:02:25 GMT

We had a pilot error with bugfix branch that accidentally performed Org
10 release (ish). This is now un-done.
If you pulled bugfix to your local clones today, do not be surprised
that part of the git history disappeared on upstream bugfix.
The problematic merge commit and a couple of hundreds commits it brought
from main are now gone from bugfix.
It is not yet the time.
https://list.orgmode.org/87v77akee5.fsf@localhost/T/#u
#orgmode

2026-10-08 19:51:03 GMT

Pedro A. Aranda, our ox-latex maintainer, is working on making LaTeX export produce accessible (PDF-UA2 compiant) PDFs. This is easy with lualatex, but not so much with pdflatex. Yet, with a good amount of automation, we might pull it off. Please help him with testing and improving the code. Announcement: https://list.orgmode.org/8733ugm0sr.fsf@localhost/T/#u
#orgmode #accessibility

2026-10-07 16:31:25 GMT

#35 bbb:OrgMeetup on Wed, October 14, 19:00 UTC+3
Another OrgMeetup is scheduled **next week**.
Earlier meetup notes:
https://orgmode.org/worg/orgmeetup.html
URL: https://bbb.emacsverse.org/rooms/orgmeetup/join
#OrgMeetup #meetup #emacs #orgmode
CC: @nprofile…9y77

2026-08-23 19:48:00 GMT

- reply

Thanks to JD Smith for contributing. https://list.orgmode.org/[email protected]/T/#t

2026-08-23 19:46:17 GMT

org-inside has been merged upstream.
This is going to be a built-in analogue of org-appear package, but less intrusive.
org-inside-mode makes it easier to edit entities with hidden contents like emphasized text, links, and super/subscripts. It alters the appearance when point is "inside" elements with hidden contents, so you know where edits will occur. The appearance is configurable, and can include face, cursor type, and unhiding ( automatic or toggled via C-c C-c).
#orgmode

2026-08-16 07:09:11 GMT

- reply

@nprofile…hgrw
TL;DR: mascot; one bug fix by Peter Mao; org-inside code review; one more patch review and two bug reports (analysis).
https://list.orgmode.org/87v79ay2ah.fsf@localhost/T/#u
Recording: https://archive.org/details/3-org-dev-meetup-on-sat-aug-16-19-00-utc-3
#orgmode #emacs #orgmeetup

2026-08-15 15:56:40 GMT

- reply

The room is open. Join: https://bbb.emacsverse.org/rooms/orgdevmeetup/join
#orgmeetup

2026-08-12 15:55:53 GMT

- reply

The room is open. Join: https://bbb.emacsverse.org/rooms/orgmeetup/join
#orgmeetup

2026-08-08 07:31:35 GMT

#3 OrgDevMeetup on Sat, August 15, 19:00-21:00 UTC+3
Org *development* meetup is scheduled next week.
The time should be suitable for US.
The meetup focuses on actual Org development, live, together.
More information: https://orgmode.org/worg/orgdevmeetup.html
URL: https://bbb.emacsverse.org/rooms/orgdevmeetup/join
#orgmeetup #orgmode #emacs #meetup
CC: @nprofile…hgrw

2026-08-07 18:28:37 GMT

- reply

@nprofile…sve2 @nprofile…fuq7 @nprofile…r0xq @nprofile…u5g7 @nprofile…wsge I would not use that paper for environmental reference. 2021 considerations are too outdated for now. I think we discussed this topic comprehensively in this thread months back when it just started.
As for practical usage, I also use guardrails and demand references (config is public). I'm also slowly moving towards local LLMs as they become more capable.

2026-08-07 18:06:52 GMT

#33 bbb:OrgMeetup on Wed, August 12, 19:00 UTC+3
Another OrgMeetup is scheduled **next week**.
Earlier meetup notes:
https://orgmode.org/worg/orgmeetup.html
URL: https://bbb.emacsverse.org/rooms/orgmeetup/join
#OrgMeetup #meetup #emacs #orgmode
CC: @nprofile…hgrw

2026-08-07 17:38:37 GMT

- reply

@nprofile…sve2 @nprofile…fuq7 @nprofile…r0xq @nprofile…u5g7 @nprofile…wsge I heard about this one before (it is viral), but did not read in the past. I have to say that it is a bit outdated and heavily debated (not all, but some claims). For example, knowledge cutoff issue is now well-addressed by RAG and search integration. Biases confirmed. I plan to read more recent papers on the topic, including https://www.lesswrong.com/posts/KWHeBG978uZuqNK6Q/hunting-undead-stochastic-parrots-finding-and-killing-the and https://arxiv.org/abs/2505.05298

2026-08-07 17:23:07 GMT

- reply

Meeting notes: https://list.orgmode.org/87cxvt7tmi.fsf@localhost/T/#u
TL;DR: ltx-talk support; exporting tagged PDFs; footnotes inside headings; hacking long paragraph highlighting; bi-directional sync between MD and Org; RFC for calling babel blocks from Elisp; Android menu; OrgDevMeetup.
Earlier notes: https://orgmode.org/worg/orgmeetup.html
#orgmode #orgmeetup #emacs
@nprofile…9y77

2026-08-03 16:17:22 GMT

- reply

@nprofile…mrfl @nprofile…7cw0 @nprofile…24wu @nprofile…muy0 Yeah. My approach is similar, although I try to get to (1) "good enough" feeling; (2) meeting Elisp standards for docstrings/custom keywords. But in this particular thread I linked to, I do not always feel good about code style, while the author is rather defensive about it. So, feeling here was rather misguiding. At least Pedro was rather annoyed by some of my comments (to my surprise).

2026-08-02 18:56:25 GMT

- reply

@nprofile…7cw0 @nprofile…mrfl @nprofile…24wu @nprofile…muy0 Hmm. It is not that I am tired or burnt. Rather wondering how much scrutiny others use for reviewing nontrivial patches. emacs-devel is infamously strict on these things, but I also understand the need to keep things stable.

2026-08-02 15:03:14 GMT

- reply

CC: @nprofile…mrfl @nprofile…24wu @nprofile…muy0 @nprofile…7cw0

2026-08-02 14:59:56 GMT

- reply

Some reference reading: https://google.github.io/eng-practices/review/reviewer/standard.html

2026-08-02 14:43:55 GMT

I am revisiting a *very* long review of a feature branch by @nprofile…qh4e and thinking about balancing between code quality, future stability, and adding significant new features. This is tricky. On one hand, I usually try to review every detail of the code and explain any issues. On the other hand, for large patches, this makes things slow, especially if I have many comments, potentially creating high contribution barrier. Any insights from fellow maintainers?
#orgmode
https://list.orgmode.org/CAO48Bk9GjRGO=gBa2HRszeSx4RjGc9-jVdHWHMMoSO4k9SMeQg@mail.gmail.com/

2026-07-28 20:06:19 GMT

I know it well why people raise defences against LLM bots on websites, but come on... do not hide RSS feeds behind those.

2026-07-28 17:18:08 GMT

- reply

Meeting notes: https://list.orgmode.org/87cxw7rqw9.fsf@localhost/T/#u
TL;DR: ox-latex survey; non-Latin PDF export; exporting tagged PDFs; buffer-locals vs Org keywords; DEADLINE syntax; Org parser in Rust; large Org files; sparse trees; anju; org-export-region-to-md; importing to Org; transient menus in Org; org-inside; canvas patch + babel = jupyter notebooks; transferring data between Emacs and browsers; LLMs (local); email providers; OrgDevMeetup
Earlier notes: https://orgmode.org/worg/orgmeetup.html
#orgmode #orgmeetup #emacs
@nprofile…9y77

2026-07-24 15:21:41 GMT

- reply

@nprofile…9y77
- There is a discussion in https://list.orgmode.org/orgmode/[email protected]/
about allowing custom links to have a AST->AST transformer that
will take effect before the export kicks in. It should allow what you
are looking for. However, the patch is not finished there, pending
some (minor) comments. If someone takes the patch and makes it into
the final shape, such feature will be come available.

2026-07-24 15:21:34 GMT

- reply

@nprofile…9y77
- ~org-export-collect-headlines~ operates on the Org AST and it
does not know anything about custom links (which are exported
to text directly, without intermediate Org/AST representation)
3/4

2026-07-24 15:21:04 GMT

- reply

@nprofile…9y77
- Sacha Chua asked a question I missed during the meetup
"is there a way an Org custom link type can mess with
org-export-collect-headlines so that #+TOC can include headings that
were exported by the custom link? (My use case: transcripts with
chapters)"
- In Org mode, what goes into TOC, is generally controlled by
~org-export-collect-headlines~. It only looks into headings.

2026-07-24 15:20:39 GMT

- reply

@nprofile…9y77 A piece of meetup notes I am writing (this meetup had so many topics, so I am taking my time). The notes are not yet done, but here is the paragraph I wrote with unanswered question:
1/2

2026-07-18 16:32:31 GMT

- reply

@nprofile…9y77 @nprofile…mnxc
TL;DR; two patch reviews, a bug fix for org-element cache, Artsiom made a small contribution to WORG, why copyright assignment is needed.
https://list.orgmode.org/87tspw1cwz.fsf@localhost/T/#u
This meetup was recorded, following numerous requests.
You can find the full recording at
https://archive.org/details/2-org-dev-meetup-on-sat-july-18-11-00-utc-3
(I recommend watching archive videos using mpv)
#orgmode #orgmeetup #emacs

2026-07-18 07:59:17 GMT

- reply

The meetup is about to start. Join: https://bbb.emacsverse.org/rooms/orgdevmeetup/join

2026-07-15 17:04:23 GMT

- reply

@nprofile…r0xq Do not forget a common rule that the result, if felt wrong, should be discarded :)

2026-07-12 07:29:53 GMT

#2 OrgDevMeetup on Sat, July 18, 11:00-13:00 UTC+3
The new *development* meetup is scheduled next week.
Note that the time should be suitable for China/Australia.
The meetup focuses on actual Org development, live, together.
More information: https://orgmode.org/worg/orgdevmeetup.html
URL: https://bbb.emacsverse.org/rooms/orgdevmeetup/join
#orgmeetup #orgmode #emacs #meetup
CC: @nprofile…9y77 @nprofile…mnxc

2026-07-10 19:06:22 GMT

- reply

@nprofile…e5lm @nprofile…muy0 By the way, I am wondering whether you thought about accessibility for emacs-reader. One of the replies on the list points that there is a lack of good screen reader that can work with PDFs on Linux, except browsers.

2026-07-10 17:23:48 GMT

We are looking to improve Org LaTeX export to produce fully accessible (PDF/UA2) documents, and we need help from users more familiar with accessibility and how to deal with it in LuaLaTeX and also PDFLaTeX.
See the mailing list call for help
https://list.orgmode.org/orgmode/87tsq6937o.fsf@localhost/
#orgmode #emacs #blind

2026-07-08 17:37:09 GMT

- reply

@nprofile…y0vl You are right that my message was not quite accurate. The numbers are correct, but I messed up the units. 10.2Wh = 10.2 * 3600 J = 36.72kJ. And the second paper was talking about J/token (obviously, W - is J/sec, which makes no sense). Now, the numbers should add up.

2026-07-08 15:44:26 GMT

- reply

The room is open. Official start in 15 minutes. Join: https://bbb.emacsverse.org/rooms/orgmeetup/join
#Orgmeetup #orgmode

2026-07-06 14:52:37 GMT

- reply

@nprofile…047x that's what I meant by "saved". If you want per-project save, need to write some hook or advice that writes to dir-locals

2026-07-06 05:59:20 GMT

- reply

@nprofile…047x What do you mean?

2026-07-06 03:58:33 GMT

- reply

@nprofile…047x AFAIK, it is saved, but not between sessions. You would probably need something custom to save it between sessions.

2026-07-05 19:09:58 GMT

- reply

@nprofile…047x yes

2026-07-05 19:07:47 GMT

- reply

@nprofile…047x compile-command

2026-07-05 16:42:20 GMT

Julien recently suggested an idea to allow links like [[pdf:<path>::(:page 12 :edges (0.240196 0.478535 0.331699 0.494949))]] in Org - specify extra information for opening the links.
Discussion: https://list.orgmode.org/[email protected]/T/#mda673691783245cba7bab64f2932f5c1520ee192
#orgmode

2026-07-05 11:15:05 GMT

- reply

@nprofile…wsge @nprofile…mx8a Could you briefly describe those ideas? (I have no idea which parts of these 2 hours worth of videos are relevant).

2026-07-05 06:36:40 GMT

- reply

@nprofile…9y77 That reminds me of the old idea I had to have short screencasts for various Org/Emacs features to put in WORG.
https://cdn.fosstodon.org/media_attachments/files/116/865/980/273/200/561/original/41594137e66880dc.png

2026-07-01 18:21:26 GMT

#31 bbb:OrgMeetup on Wed, July 8, 19:00 UTC+3
Another OrgMeetup is scheduled **in 1 week**.
Earlier meetup notes:
https://orgmode.org/worg/orgmeetup.html
URL: https://bbb.emacsverse.org/rooms/orgmeetup/join
#OrgMeetup #meetup #emacs #orgmode
CC: @nprofile…9y77

2026-07-01 18:12:21 GMT

- reply

Meeting notes: https://list.orgmode.org/orgmode/878q7uzip7.fsf@localhost/
TL;DR: meetup recording; ox-latex patches; Org mode communities; typst; contributor liaisons; org-inside; metadata in properties; org-capture-ref, zotero magic button; org-noter, mobile notes; reading research papers; emacs-reader; canvas patch; Linux+Emacs system theme: tangle&babel; fixing ob-ruby; resources for beginners; org-social; Org grammar parsing; cloning mailing list.
Earlier notes: https://orgmode.org/worg/orgmeetup.html
#orgmode #orgmeetup #emacs

2026-06-26 18:44:06 GMT

Finally finished watching EmacsConf talks (all of them). Very slooowly :)
P.S. EmacsConf website is nice to view in eww + shrface

2026-06-22 18:56:11 GMT

Meetup summary:
Turned out more like a live-stream (over 2 hours), with people able to join by voice.
I was hoping to get more people trying to code. An earlier smaller trial did yield a patch, but not here (even though we had more people).
Next time: probably ask newcomers to introduce themselves, try coding. Maybe record the stream. And maybe float the time slot across Saturdays to cover more timezones, like Australia/China.
More: https://list.orgmode.org/8733ygi64t.fsf@localhost/T/#u
#orgmeetup #orgmode @nprofile…9y77

2026-06-21 18:00:47 GMT

- reply

And to be clear, the liaisons are helping a lot. Many of the bug reports and patches were initial reviewed or even applied before I got to them. I also did not have to handle literally everything - giving a 2 week time on easier patches to be reviewed. On the other hand, more activity with reviews and responses = more contributions. Maybe we will need more liaisons :) #orgmode

2026-06-21 14:52:23 GMT

I think I reviewed all the pending patches so far. The last month was a combination of me being unavailable/too tired on weekends after the work trip and a spike of submission after contributor liaisons joined.
Now, 130-ish emails to go. Non-patches. Slowly catching up.
https://cdn.fosstodon.org/media_attachments/files/116/788/651/286/352/592/original/044b13be5162f173.png

2026-06-20 15:50:40 GMT

- reply

Starting now. Note that there was a typo in the join link on the web page (will be fixed after next rebuild). Correct link: https://bbb.emacsverse.org/rooms/orgdevmeetup/join

2026-06-17 03:33:51 GMT

- reply

@nprofile…mnxc @nprofile…9y77 Yeah. But if I move earlier, it will not be so good for bay area. One option could be rotating the time

2026-06-16 19:37:53 GMT

New Org*Dev*Meetup on Saturday: let's contribute to Org mode together, online
I want to try doing a development-focused meetup this weekend:
1. I will work on recent bugs
2. People can ask me anything or comment on the code or development
3. If someone wants to try, we can quickly find a fresh bug worth fixing, and get the code done together
More info: https://orgmode.org/worg/orgdevmeetup.html
#orgmeetup #orgmode @nprofile…9y77

2026-06-10 15:53:26 GMT

- reply

The room is open. Join: https://bbb.emacsverse.org/rooms/orgmeetup/join

2026-06-06 17:22:39 GMT

"Fun" with btrfs, snapper, and balancing requiring extra space. Oh, well.

2026-06-06 08:30:55 GMT

I believe that I finally managed to track down the annoying element cache bug that is linked to editing plain lists. It has been reported a number of times, but it is hard to reproduce. Recently, Jason Terk has managed to record detailed logs of what is hapenning when the bug is trigerred, and I think the reason is related to successive edits coming between cache syncs. Now, if we confirm that https://list.orgmode.org/orgmode/CAKcq1chJuVKb7C=vYWN9jwKa=Yr_SC6x9S3Mq04TH0jGgjkriw@mail.gmail.com/ really works, it should hopefully eliminate one of the last cache bugs.

2026-06-04 12:29:15 GMT

- reply

@nprofile…sve2 @nprofile…863e @nprofile…r0xq @nprofile…u5g7 Yeah. But opening the report they use GPT-4o energy numbers (Figure 11), which are an order of magnitude higher than the newest models. Also, they project 40% datacenter energy consumption to be used by LLMs in 2030 - it is actually *better* than I expected (LLM consume less than traditional datacenters).
The report raises valid points about data center concentration and overuse of resources, but that's what we discussed. I do not see much new info.

2026-06-04 06:02:26 GMT

#31 bbb:OrgMeetup on Wed, June 10, 19:00 UTC+3
Another OrgMeetup is scheduled **next week**.
Earlier meetup notes:
https://orgmode.org/worg/orgmeetup.html
URL: https://bbb.emacsverse.org/rooms/orgmeetup/join
#OrgMeetup #meetup #emacs #orgmode
CC: @nprofile…9y77

2026-06-03 18:07:09 GMT

- reply

@nprofile…sve2 @nprofile…f7nc @nprofile…r0xq @nprofile…u5g7 Well. There are. Simple example - local model for translation that allows communication with non-English speakers online. That's a pure improvement over the existing non-private cloud services.

2026-06-03 17:58:36 GMT

- reply

@nprofile…f7nc @nprofile…r0xq @nprofile…sve2 @nprofile…u5g7 Including open source LLMs trained by research institutes? Sorry, but this sounds like a false dichotomy.

2026-06-03 17:55:26 GMT

- reply

@nprofile…45cc @nprofile…wsge @nprofile…eg75 This works to a point. You can ask 2X, 3X, but when you start asking 1000000X, people will laugh at you and ignore.

2026-06-03 05:03:57 GMT

- reply

@nprofile…wsge @nprofile…eg75 One of the speakers mentioned that all the resources should be channelled to fight the climate change (evena pausing the usual development). This makes me wonder what kind of resources are *already* dedicated to fight climate issues. I know that in EU there is a lot of focus on recycling, sustainability, and green energy. China is going green energy big time. Any other? What makes the biggest impact in practice? (IMHO, 100% dedication is naive; it won't happen).

2026-06-01 20:23:06 GMT

- reply

@nprofile…863e @nprofile…sve2 @nprofile…r0xq @nprofile…u5g7 After this discussion, I still think that the total consumption is not as problematic. Regional issues are real, as confirmed by other participants on this thread. I am hopeful that they can be solved with green regulations for data centers. (I have 0 idea about US case though. That place is crazy).

2026-06-01 20:19:32 GMT

- reply

@nprofile…863e @nprofile…sve2 @nprofile…r0xq @nprofile…u5g7 Then, I'd like to refer to https://www.iea.org/reports/energy-and-ai/energy-demand-from-ai#abstract again:
"... data centre electricity demand growth accounts for less than 10% of global electricity demand growth between 2024 and 2030 in the Base Case. Other key drivers, such as industry output growth and electrification, the deployment of electric vehicles, and the adoption of air conditioning, lead the way. ,,, data centres... tend to concentrate in specific locations...."

2026-06-01 20:02:41 GMT

- reply

@nprofile…863e @nprofile…sve2 @nprofile…r0xq @nprofile…u5g7 you missed my next message there.

2026-06-01 17:43:33 GMT

- reply

@nprofile…sve2 @nprofile…fuq7 @nprofile…r0xq @nprofile…u5g7 @nprofile…wsge I'd like to point out that I absolutely agree that current situation (especially in US) with data centers is not particularly sustainable. Especially as they concentrate in the states with insufficient resources. But I do not agree that it is about LLMs. It could have been the same with any new tech. It is not LLMs to blame, but the government that does not care about ecology.
Why not building data centers where the resources are available?

2026-06-01 17:39:09 GMT

- reply

@nprofile…sywp Sorry, I am lost. The state has laws that are not particularly environmental-friendly. But how does it have anything to do with other places that regulate things better and do not hurt the citizens?
AFAIU, your position is that LLMs should be banned or limited everywhere. What I am trying to show is that it is not about LLMs. If state does not care about ecology, there will likely be some tech companies that profit on that. But it does not mean that underlying tech should be banned.

2026-06-01 17:35:15 GMT

- reply

@nprofile…863e @nprofile…sve2 @nprofile…r0xq @nprofile…u5g7 But considering that $$$ costs are also scaling exponentially, I believe that slowdown is inevitable (LLM companies are already half of global capex). Especially given that the theoretical information enthropy limit is within the reach: https://www.youtube.com/watch?v=5eqRuVp65eY

2026-06-01 17:31:37 GMT

- reply

@nprofile…863e @nprofile…sve2 @nprofile…r0xq @nprofile…u5g7
That said, https://epoch.ai/trends#data-centers shows that training costs do grow exponentially so far. According to report I previously cited (https://www.iea.org/reports/energy-and-ai/ai-and-climate-change), despite the grows total energy usage for LLMs is still smaller than traditional data centers (and, combined, still very small vs total energy usage). I won't be surprised if it exceeds them though.

2026-06-01 17:28:24 GMT

- reply

@nprofile…863e @nprofile…sve2 @nprofile…r0xq @nprofile…u5g7 I have reviewed recent literature on the topic. Most people cite https://arxiv.org/abs/2204.05149. That article claims that inference energy consumption is larger than training. The article is from google, but I saw similar claims from research publications. Numbers vary, but the inference dominate (>50%) in all the claims I found in papers. So prompt energy cost does matter.

2026-06-01 16:23:58 GMT

- reply

@nprofile…sywp I did not mean that data center developers are right or that they had no other choice. They do.
Certainly, I do not call these developers "victims". They are not. What I am saying is that elsewhere, where the state cares more about ecology, the impact of data centers is much, much smaller.
In other words, ecological damage of data centers in not inherent to LLMs, but depends on state policy for ecology (not LLM policy).

2026-06-01 03:11:04 GMT

- reply

@nprofile…sywp So, should we ban all the data centers across the world just because some bad actor state makes them profitable at the expense of environment? My answer is no. We need to look into broader world context.

2026-06-01 03:09:54 GMT

- reply

@nprofile…sywp There are places were data centers are built in sustained way. For example, connecting cooling with city heating system in winter. The impact is much smaller. 2/3

2026-06-01 03:07:28 GMT

- reply

@nprofile…sywp That's not what I meant. My point is different:
There is a state that does not enforce environmental standards of data center operations. Of course, that's cheaper - data cennters do get built there. More and more. Exactly because there are no regulartions. And, of course, the state ecology suffer. But does it have to be that way? 1/2

2026-05-31 20:03:14 GMT

- reply

@nprofile…sve2 @nprofile…fuq7 @nprofile…r0xq @nprofile…u5g7 @nprofile…wsge Well. The reality is that we do not have common guardrails. The best is per-country basis. Nothing global and binding at the same time. As the recent years showed, we have no reall way to enforce international agreements.

2026-05-31 19:54:48 GMT

- reply

@nprofile…sve2 @nprofile…fuq7 @nprofile…r0xq @nprofile…u5g7 @nprofile…wsge For that state, yes. And for some US states that even downgrade the environmental regulations to adapt more datacenters, it is certainly terrible. But I see no point blaming LLMs for the more general lack of good environmental policy (which goes beyond just datacenters). In the contex of the world, imapct of LLM datacenters is much smaller.

2026-05-31 19:44:53 GMT

- reply

@nprofile…fuq7 @nprofile…sve2 @nprofile…r0xq @nprofile…u5g7 @nprofile…wsge Yes, you are absolutely right. The link I provided shows exactly this. My point is the scale - I do not yet see GPU datacenter surpassing conventional datacenter capacity. And GPU + conventional combined are still a small fraction of total energy consumption. Yes, it is still an increase in carbon footprint, but nowhere near the major contributors.

2026-05-31 19:00:56 GMT

- reply

@nprofile…sve2 @nprofile…r0xq @nprofile…u5g7 To be clear, LLM energy consumption is real. @nprofile…wsge referenced https://www.iea.org/reports/energy-and-ai/energy-demand-from-ai#abstract that clearly shows that GNU datacenter buildup is comparable to conventional datacenters. But it is not as dramatic as often painted.

2026-05-31 18:42:34 GMT

- reply

@nprofile…sve2 @nprofile…r0xq @nprofile…u5g7 More specifically, an average CI run consumes ~37kW [1], while newer LLMs consume about 0.05-0.1W/token [2], which means that a CI run is an equivalent of burning ~0.5M LLM tokens. And LLMs keep improving in efficiency, unlike the CIs. I would worry more about CI honestly.
[1] https://arxiv.org/html/2410.15816v2
[2] https://arxiv.org/html/2604.09048v1

2026-05-31 18:22:20 GMT

- reply

@nprofile…sve2 @nprofile…r0xq @nprofile…u5g7 Hard to say. Running something like llama is very easy these days.

2026-05-31 18:20:43 GMT

- reply

@nprofile…sve2 @nprofile…r0xq @nprofile…u5g7 Yes, I heard these arguments. So, did you, for example, research about the numbers for electricity use? My own research showed that a complex LLM prompt uses less electricity that a CI run on a cirtual machine. (on the latest models). I think the water and noise will go similar if we compare normal data centers and GPU ones.

2026-05-31 18:18:43 GMT

- reply

@nprofile…r0xq @nprofile…sve2 @nprofile…u5g7 Right. It is mostly that I do not like the blind critisism that makes blind assumption about LLMs. The situation is more nuanced.

2026-05-31 18:16:34 GMT

- reply

To be clear, there was no bad intention behind the vibe-coding. But a totally unexpected behavior of Claude code.

2026-05-31 18:15:17 GMT

- reply

@nprofile…sve2 @nprofile…r0xq @nprofile…u5g7 BTW, I keep wondering about the data the argument about "destroying the planet" is based upon.

2026-05-31 18:13:53 GMT

- reply

@nprofile…sve2 @nprofile…r0xq @nprofile…u5g7 Please consider that not all LLMs are made by big tech. There are LLMs that are open source and LLMs that can run locally.

2026-05-31 18:12:16 GMT

- reply

@nprofile…r0xq On a related note, we had a whole package vibe-coded by LLM based on unfinished patch from the mailing list. What LLM did is crawled the mailing list archives (without explicitly
being asked to do so), found the unfinished patch, and basically reused
its implementation to write the code. It even documented the process in
https://gothub.r4fo.com/jwiegley/pending/blob/main/RESEARCH.md The code
itself is not copied, but many of the design decisions, and even some
jokes (!) from our patch were reused in the LLM version.

2026-05-31 18:00:57 GMT

- reply

@nprofile…r0xq @nprofile…u5g7 I suspect that changelog changes could be considered trivial

2026-05-31 18:00:31 GMT

- reply

@nprofile…r0xq @nprofile…u5g7 I do not think that commit message if a subject of copyright. Changelog might be.

2026-05-31 17:52:17 GMT

- reply

@nprofile…r0xq @nprofile…u5g7 Yes, except the README part of the changes (unless they are trivial).

2026-05-31 17:27:38 GMT

- reply

@nprofile…r0xq I think you do not really need to delete the post. There are our replies just below with @nprofile…nlhz pointing to worg page.

2026-05-31 17:12:55 GMT

- reply

@nprofile…r0xq Note that people reading it and who never tried org-capture may simply be reluctant to try after reading your and similar posts. So, I really do not find such complaints helpful for others. Although I do understand the urge in similar situations.

2026-05-31 17:10:57 GMT

- reply

@nprofile…u5g7 @nprofile…r0xq In Org mode, we have a vide-coded patch that re-implements org-table. The patch has issues and we went through several iterations of vibe-coding and vibe-reviewing, and I am not realy convinced that the current state is acceptable. However (!), that library is very old and hard to maintain, with legacy bugs and undocumented behaviors. Further, nobody is dare to work on it (it is a giant job). So, vibe-coding can be better as a kick start.

2026-05-31 17:04:11 GMT

- reply

@nprofile…r0xq The current guidance to GNU projects (including Emacs) is not to accept LLM contributions until GNU issues a public policy guideline. This is to avoid potential copyright implications. The only possible exception is small trivial patches.

2026-05-31 16:46:53 GMT

- reply

@nprofile…r0xq as it usually goes, "not working" is never a useful report.

2026-05-31 12:47:58 GMT

- reply

@nprofile…wsge @nprofile…mx8a @nprofile…u622 In practice though, implementing optimal society is itself taking time and efforts. The stable and good end state may exist, but the way there may neither be easy nor guaranteed (without trying hard).

2026-05-31 12:46:41 GMT

- reply

@nprofile…wsge @nprofile…mx8a @nprofile…u622 The end of the essay indeed offers the philosophical solution. It is also not inconsistent with the optimal behavior within game theory example you suggested - playing optimally in non-zero sum game may not only outperform, but also be stable against bad actors. That require certain structure of society.
1/2

2026-05-31 06:45:58 GMT

- reply

@nprofile…9wat @nprofile…u622 @nprofile…wsge This reminds me of excellent writeup on why it is so important to have regulation (by law, or society norms) https://www.lesswrong.com/posts/TxcRbCYHaeL59aY7E/meditations-on-moloch