@DavidCarlisle I am sure your fix works. I did not get chance to try it as I ended up finding and fixing my program so it does not generate an empty dmath environment by mistake to avoid this problem in first place. But thanks for the patch/fix.

it's more important to test it does not raise spurious errors in good documents, so if you could test it it would be helpful

@zetyty I have a duck one /ducks

@Nasser What is this fix for?

@FaheemMitha See this horr^H^H^H^Hbeautiful fix.

@mickep elegant code, like its author

@DavidCarlisle To be honest, I laughed when I saw the code. I am impressed that humans can write and understand it.

@mickep as I indcated to Nasser, I don't fully understand it. that macro is documented as a recursive call and \tracingall showed it was looping {and box 2 was void) so easy to add a voidbox test and force an error and the loop to end, but I'm not 100% sure what it's doing and whether there is a possibility of box2 being void in cases that wont loop and should not error

@DavidCarlisle I see. So maybe others documents will now break. Hopefully not.

Time for you to rewrite breqn in modern language? People seem to like it's features...

@Nasser I tested your MWE. Yes, it hangs at 100% CPU. Have you reported it as a bug, or not?

@mickep which is why it would be good if @Nasser would test his 100000 page breqn document.....

@mickep modern language, like rust ?

BTW, I started up Zoom as a desktop program some time ago. At least a week. Possibly 2. I just noticed that it is running at 100% CPU with multiple processes. Had to do quite a lot of killing. Nasty. I was told by someone on U&L Chat to avoid the Zoom program on Linux. I'd say they have a point.

@DavidCarlisle Yes, that is a simple (well) compilation away.

@DavidCarlisle You are free to choose your favorite.

@FaheemMitha it has been working fine for me during the pandemics. But why keep it open for so long?

@mickep I didn't. It's not open, AFAIK.

But the processes were still running. Regardless 100% CPU with no activity is not a good look.

@mickep I have known breqn a long time, Michael sent me pre-releases, I was working on bm at the time and we looked in to making them work together (but gave up;-)

@Nasser github.com/wspr/breqn/issues, but you already posted one issue, so you know where it is.

@FaheemMitha although in practice as likely to reach a maintainer here

@Nasser I apologise for being a busy-body. I don't even use breqn.

@FaheemMitha I think it by default keeps running. So, one has to quit manually. But of course it here sounds like something has gone wrong...

@DavidCarlisle I see. So now you have the chance to combine them and write a very appreciated packare.

@mickep do you have anything to help linebreak display math in lmtx?

@mickep Regardless, 100% CPU for something that isn't doing something isn't reasonable. Usually the sign of a program that isn't working properly.

@Rmano Of course, I should have guessed^^

@Rmano Maybe I'll use it for my meeting with my boss to convince him to use LaTeX for our company, (it will show that, contrary to appearances, LaTeX can be fun!). It's probably the small details that will make a huge difference^^

@Rmano feature request: allows miniframes with little ducks in places of navigation bullet symbols^^

@DavidCarlisle Yes, penalties (not set up completely yet). Remember that they are done as paragraphs there. So multiline is automatic.

@zetyty easy to add:

\documentclass{beamer}

\usepackage{tikzducks}

\useoutertheme{miniframes}
\setbeamertemplate{mini frames}{\tikz{\duck[scale=0.08];}}
\setbeamertemplate{mini frame in current subsection}{\tikz{\duck[scale=0.08,body=lightgray];}}
\setbeamertemplate{mini frame in other subsection}{\tikz{\duck[scale=0.08,body=lightgray!50];}}

\begin{document}

\section{title}
\begin{frame}
	abc
\end{frame}
\begin{frame}
	abc
\end{frame}

\section{title}
\begin{frame}
	abc
\end{frame}
\begin{frame}
	abc
\end{frame}

@UlrikeFischer blame you? tex.stackexchange.com/a/689524/1090

@DavidCarlisle hm. If I comment the last 80 lines or so from luapstricks it behaves again.

@UlrikeFischer vvv

\documentclass{article}
\makeatletter
  %     pgf/tex/generic/pgf/systemlayer/pgfsys.code.tex
  \directlua{
  local lft = lua.get_functions_table()
  lft[\string#lft+1] = function()
      local lhs = token.scan_string()
      local rhs = token.scan_string()
      if lhs < rhs then
          tex.sprint(-2, "-1")
      elseif lhs == rhs then
          tex.sprint(-2, "0")
      else
          tex.sprint(-2, "1")
      end
  end
  token.set_lua("pgfsys@strcmp", \string#lft, "global")
  }


\typeout{^^J!!!!1: \meaning\pgfsys@strcmp^^J\pgfsys@strcmp{aa}{bb}}

@UlrikeFischer @JosephWright ltluatex bug I think

@UlrikeFischer luatexbase.new_luafunction uses e@alloc@luafunction@count but should use size of lua.get_functions_table()

@DavidCarlisle is pgf doing this correctly? luapstricks uses luatexbase.new_luafunction, but pgf seems to manipulate the table directly.

@samcarter Wonderful!! The little gray ducks are so cute!

@UlrikeFischer i think it was an error to allocate these via a tex counter, but having done that i think it should be kept in sync with the table size. I added to the answer

Who thinks it would be a good idea to create a TeX community on Codidact? It would be an interesting alternative for people who like SE style but would like to have more control over the evolution of the site. However, such a site cannot exist without the help of people like you. This involves starting from scratch but with the opportunity not to make the same mistakes (even if some here never make mistakes^^).

PS: as for the duplicates with TopAnswer, it is undeniable but the two sites will perhaps be quite complementary in terms of "user profile" and they will probably be strongly linked.

PPS: if this discussion has already been conducted here several times with the same conclusion, could you tell me and I will not insist (well not too much ;-)

@zetyty I think the challenge you face is that bar some very minor things about appearance, the Powers basically leave us alone here anyway - I think you'd need to articulate clearly what you feel would be different

@zetyty I think we have had this discussion multiple times. The main consensus seems to be that TeX.SX has been well established for users to come to and it has gained enough momentum to have knowledgeable people in all aspects of TeX here. The hard part about restarting somewhere else would probably be rebuilding the community, not the content (after all, the CC licenses would allow for some sort of import).

Personal opinion: I would want any alternative platform to federate (in the fediverse-sense). Unfortunately, the current alternatives lack seriously in that regard.

@zetyty The ducking miniframes would be ideal for a talk about en.wikipedia.org/wiki/The_Ugly_Duckling :)

@JosephWright Do you want to say (i) that I would have to bring you arguments in favor of Codidact (from a community power point of view) or (ii) do you just want me to ask myself the question clearly to be sure there is really something to gain and not just appearences?

If you invite me to make point (i) then I will have to politely decline your request because I do not feel legitimate to argue in front of people like you.

I just want to know if any of you would be willing to start from scratch in a new community with the hope of having more power over the new features.

*When I say "in front of people like you" that mean in front of people who participate activaly to what LaTeX is today since years!

@zetyty I meant that if you want people to choose another site on the basis of 'more control over the evolution of the site', then being able to give concrete examples of the outcomes would be good

@zetyty In TeX terms the content presumably would be similar (or at least of similar quality), so it's the non-TeX aspects that are important

@zetyty For example, I also have an account on TopAnswers and I can say what's different, but I don't think that (at present) there are enough benefits to the day to day user to lead to a wholesale switch

@zetyty I have in the past been very active on c.t.t. and on latex.org, but the balance of features on TeX-sx plus a general move of people means nowadays I'm not

@JosephWright WRT to your message on #1092, does this mean that this issue is not likely to be fixed soon? If so, the dev version is likely to remain unusable for me till that happens.

I didn't entirely understand your message, but it sounds like you bundle together fixes before applying them. Not one at a time.

Also, I don't know what PRO is.

@FaheemMitha 'Soon' is a movable feast

If anyone cares about what I'm talking about, github.com/latex3/latex2e/issues/1092#issuecomment-1606224247

@FaheemMitha PR0 is 'pre-release 0' - it's the dev. release we make when we do a full LaTeX release and therefore is identification to the main release other than the version string

@JosephWright Feast?

@FaheemMitha A movable feast is something that doesn't have a fixed date

@JosephWright Oh, that was a zero. Sorry.

@JosephWright Feast is normally used for things to eat. Like a banquet. Generally unlimited things.

@JosephWright -- One not-so-minor disruption was the effective disabling of site-specific ads, which meant that we are no longer able to do things like announce TeX conferences. These have been replaced by income-producing (for the Powers) ads that are often distracting and irrelevant to the TeX community. While it's true that any site like this needs financial backing, the change could have been made more gently.

@JosephWright Are you and the other mods not concerned about the AI thing? While I don't pretend to understand the implications, it seems to have much of SE up in arms.

@FaheemMitha Yes, that's entirely true: here, a 'moveable feast' is referring to the date of the feast, not the meal itself. It's used in English to mean 'something without a fixed date'

@JosephWright I'm not familiar with that expression.

@barbarabeeton Was that a policy decision made by the company that owns the site?

I did find en.wikipedia.org/wiki/Moveable_feast

Also a memoir by Hemingway.

@FaheemMitha That's where the use of the term comes from

@FaheemMitha dictionary.cambridge.org/dictionary/english/movable-feast

@JosephWright Ah. Well, I could not find any reference to the use of the general term.

@JosephWright OK. Thank you very much.

@zetyty Please don't think I'm being negative: I just thing that some examples of where a different owner/structure would make a difference would be useful

@zetyty Codidact does not have a chat site, as far as I know.

I could be wrong. I'm not a user there.

@zetyty It's true that TeX SE seems to have strong visibility in the TeX world. It would be a shame if it were to go away. That's isn't the case for most of the sites on SE, as far as I know. In that sense it is an outlier.

@FaheemMitha -- What I know is from how the change played out. It happened just as we were starting some date-specific announcements, and then told that we could post announcements only during a very short period at the beginning of the year, and then they couldn't be changed until the next year, making the option totally useless. We used to have very useful and attractive TeX-related ads.

@barbarabeeton Oh, that's too bad. They want to use the space to make money, by the sound of it.

With respect to that patch issue, I may be misunderstanding what dev means. Can anyone point me to a relevant link which documents the relationship between these two branches?

@JosephWright -- Another change/loss was the discontinuation of a site-specific blog, which was used largely for interviews with TeX people. The interviews were salvaged, posted elsewhere, and linked from the TUG interviews page. But that has a different, likely smaller, audience.

@FaheemMitha At least two things. On one side, SE users and mods are concerned about AI generated not-high-value content, spamming the site quickly, hard to curate/edit. On the other side, people can now use ChatGPT etc. to get AI-generated answers from sources such as SE without going to SE, so fewer page views and less advertisement revenue for the company that acquired SE for about 2 billion dollars.

(their investment is now the interface between content producers, us, and AI crawlers, less humans soon, why should they keep financing the infrastructure and staff)

@zetyty personally I can't see it being useful (I don't use topanswer either). The specifc site mechanisms here were different enough to more or less kill off comp.text.tex and be the main help forum for the community. Alternatives so far have minor technical differences but not enough to move the community so lead to fragmentation or duplication and I just don't see the advantage.

@barbarabeeton Saved on tex-talk.net/category/tex-sx/interviews , admitted that it's not much promoted/linked, this could have been improved and still can

@barbarabeeton Again, i think you've got to be quite hard-headed - how many of the total visitors to TeX-sx looked at that? We are back with 'active' users/developers, a very small number of people who likely could use another mechanism. Plus, on a blog, it's better to ask for a static link from a Q&A site to a blog hosted on a dedicated system

@StefanKottwitz -- Thanks for the reminder.

@barbarabeeton I remember those interviews.

@barbarabeeton oops you remind me I have a re-interview form to fill in....

@StefanKottwitz Do you think AI (a misnomer, if there ever was one), will do better than a regular chat engine?

@JosephWright -- But there's no link in any obvious place. Where could it be placed so that a random reader could notice it?

@FaheemMitha -- AI is only as good as the information available for a basis. And there's so much misinformation out there about TeX, and it's hard for an "uninitiated" observer to tell whether it's good or bad.

@barbarabeeton I also don't see why AI would understand TeX any better (i.e. not at all), than a search engine.

Though from what I've read it does surprisingly well at answering even not so basic questions. I don't really understand how that works. I would expect a mish-mash of vaguely relevant stuff punctuated by gibberish.

@FaheemMitha -- I think you've got the concept.

@FaheemMitha -- Except that "facts" are presented in frighteningly proper English (or presumably whatever natural language). So it sounds scholarly, even if the content is total nonsense.

@FaheemMitha I guess you haven't used chatgpt

@DavidCarlisle No, I haven't. Have you?

@DavidCarlisle @FaheemMitha's sounds about right: vaguely plasable at first reading, rubbish when you get into the detai

@FaheemMitha yes

@JosephWright popular content may be ok, specific niche technical content not yet, but main SE/SO visitors' searches may be answerable by an AI bot that had access to the sources

@FaheemMitha Note that ChatGPT is in fact exactly a chat engine. It is an AI-powered chat bot. Nothing more and nothing less. It has no access to the live internet (yet) and it can help in coding in certain situations obviously, but it is not primarily trained to do so. Its output is optimized so that it sounds well. But it does not necessarily need to be true. I think there may well be the possibility that someone creates an AI that understands TeX. ChatGPT is not such an AI.

@JasperHabicht Yes, it's a language model, statistically. It can be trained on English or German or Javascript or TeX, the latter wasn't the main focus I guess :-)

Random fast check:

@StefanKottwitz The problem is exactly that it is a language model. In order to really understand a programming language, I would assume that it also has access to the defintions of the variables and functions. And this is probably then not really a language model anymore ... but I might be wrong

@StefanKottwitz This is of course a very easy task since such a defintion is very likely to be found in the large corpus ChatGPT has been trained with.

People can ask simple questions without being patronized, question closed as duplicate, asked for a minimal example, whatever.

It's probably ok for easy stuff (80%?) people usually used google (arrive at SE and "pay" with advertisement view) and now they may not come to SE if the bot shapes a good answer for a part of topics

@JasperHabicht yes, it's a "stupid" bot working on a huge amount of data

@StefanKottwitz It would actually be interesting to compare ChatGPT answers to questions triggering @barbarabeeton's beginner mistake talk at TUG to check whether it's really good for the easy stuff.

\begin{center}\begin{figure}[H]...\end{figure}\end{center}

@StefanKottwitz Which is ok. Yes, right. It can help people who are unable to process google results to google things =D

@yo' To be fair, I've seen quite a number of “natural intelligences” write code like that :D

\begin{center}\begin{equation*}...\end{equation*}\end{center}\vspace{2mm}\\\textit{Something that should have been a paragraph-level heading}

@TeXnician And since a coputer is only as intelligent as the person who sits in front of it ... why should ChatGPT do any better ...

It can generate templates, code snippets, and it corrects errors in code. (adds an environment around \item etc)

@TeXnician well, fair point. And the second piece of code fails with an error...

it's no replacement for a human help (yet) but it's a mechanical assistant with a knowledge base and engine

you get a better idea of the technology if you ask about python rather than tex, as it has been trained on a lot more python. But last year such systems were not generally available, this year they are but give easily spotted answers, in 10 years they will probably be more accurate than most humans

@StefanKottwitz But it does so, because of the corpus it has bee trained with. It doubt that it does so because it understands, for example, how an itemize environment is defined

@samcarter forwarded to me an answer from ChatGPT to the request "Tell me about Dante e.V." I don't have permission to share the response, but it's truly amazingly literate (and literary) garbage. I suspect similar nonsense would result from questions about TUG, GUTenberg, NTG, etc. Totally "lost in translation".

@JasperHabicht it learned the syntax and adjusts to the situation, which is probably ok for usage

@StefanKottwitz Yes, I do think so as well! And as I said: AI will surely be able to handle this. But you always need to look at the aim a certain AI has been created for. It is "ChatGPT", so it has been named for a reason like that ...

@barbarabeeton specific knowledge can be garbage and made up, especially as it has limited data, and missing data is the point here (plus confusing) but that can be improved

Somebody recently asked on the main site whether there would be any AI-supported packages for (La)TeX. I thought that one could probably create some AI using Lua ... but then I was not sure how this could help with typesetting exactly =)

@barbarabeeton vvv

User
Tell me about Dante e.V.
ChatGPT

I'm sorry, but as of my last knowledge update in September 2021, I don't have any information about an organization called Dante e.V. It's possible that Dante e.V. is a relatively new organization that was established after my knowledge cutoff. Without further context, it's difficult for me to provide specific details about Dante e.V.

However, in general, organizations with "e.V." in their name are commonly found in Germany. "e.V." stands for "eingetragener Verein," which translates to "registered association" in English. Associations (Vereins) in Ger…

@barbarabeeton I guess it can be used as an inspiration (bring up topics/infos) and a re-processing assistent (translator, re-writer, summarizer)

@barbarabeeton not bad

@TeXnician -- No, just unschooled. The problem is, the bad code appears too often, and how does an AI processor know what is good and what is bad, if the corpus hasn't been curated?

@barbarabeeton The AI has been trained to chat primarily. Not to code, as far as I know. Therefore, the error tolerance has not been set too high. This, at least to my understanding, is the reason that it outputs nonsense sometimes in order to at least output something and continue the chat. Of course, in programming, even small errors are ... bad

Dante e.V. is the German user group for the TeX typesetting sytem
ChatGPT

Thank you for providing more context. Dante e.V. is indeed the German user group for the TeX typesetting system. TeX is a widely used typesetting system developed by Donald Knuth, primarily used for creating high-quality documents with precise typography. It is especially popular in the academic and scientific communities for the creation of research papers, books, and technical documents.

Dante e.V. (Deutschsprachige Anwendervereinigung TeX e.V.) is an association that supports and promotes the use of TeX and relat…

@barbarabeeton Well, the AI corpora are usually assumed to amass a sufficient amount of documentation/curated knowledge to counter misinformation. Unfortunately, the sourcing of this knowledge is usually website-based which is not a useful domain assumption everywhere.

@barbarabeeton ^^ not bad either

... and this is what I meant in the beginning. An AI to help you with coding would better output: "sorry I don't know" if it is not 100% sure about the output. But ChatGPT does not do that, because it is a chat bot.

I think ChatGPT is not specialized enough for such tasks as helping to code with TeX

@DavidCarlisle -- Somebody has helped it along in the meantime.

@JasperHabicht Well, it does output warnings if you ask it about certain things it considers bad (violence/hate etc.). Fostering misinformation is not something the ChatGPT developers consider bad ;)

If AI chatbot answers are good enough for general Javascript, Python, Travel, Photography, Home Improvement, Gaming etc. answers in 80% of cases, it's a serious threat to the 2 billion dollar traffic-advertisement model and if that model fails, all can go down with the infrastructure. Except if SE provides a better AI approach, I guess they try.

@barbarabeeton you don't get the same answer if you ask the same question twice, and if it was inserting Italian references it had probably picked up Italian context earlier in the chat

@TeXnician But that it ouputs a warning for hate/violence ... is more because it knows that some words are bad. Does it output a warning if you feed it some code for improvement? If no, this is bad UX, because the developers should be aware (and therefore tell the user) that ChatGPT cannot fully solve such tasks and it is probably a good idea that the user test their code

@DavidCarlisle -- No, the only Italian "reference" was "DANTE".

:63848907 If your webcam was on, it even could probably see what you were eating while asking the question ... rosted duck? pineapple pizza? oh no!

:63848907 -- I believe in Germany.

@barbarabeeton sorry yes was thinking you said carlatex, but anyway point is I think policy of banning ai answers is doomed as you can only spot them while they are bad and tha is just a temporary development phase

I assure you, if I have a medical question, I'm going to ask an M.D. with whom I'm familiar. I've gotten enough wrong answers from them (later remedied, thank goodness) that I'm not going to trust an AI bot that I don't know how it was trained.

@StefanKottwitz I think it is a good tool. But the users need to learn how to use it, and know about what it can do and what it can't ...

@barbarabeeton But how do you know that the physician does not use ChatGPT behind their desk ...?

@barbarabeeton you will, or at least the next generation will. systems for filtering scans and the like already using custom ai model for spotting whatever they are scanning for

@DavidCarlisle -- Yes, and those have been trained on curated data. That makes a difference.

@JasperHabicht -- I know my primary uses Internet lookups. But he also knows my history, and can therefore filter out irrelevancies.

@barbarabeeton sure but "ban ai" policies cover a massive but general model like chatgpt or a custom models someone may make from c.t.t or ctan or whatever, in the end the models will get to be good enough

@DavidCarlisle I think (but maybe I am alone): If an AI-generated answer is really good, why not allow it? The current problem on SE is that these answers are often wrong. But answers of humans on SE are also not always right. So, maybe currently it might make sense to ban AI-generated answers, because this runs against the spririt of SE sites, but ... I think it is neither feasable nor makes it sense to ban them in the long run. If such sites become obsolete, because great AIs exist, well so be it.

@JasperHabicht yes agree

@DavidCarlisle I actually don't think that sites like SE will ever become obsolete. Someone at least has to discuss about how to train/contain/deal with AIs, right? =)

@DavidCarlisle -- We can hope. Special-purpose tools, based on a known and validated corpus, have been demonstrated to be reliable. But the new "public" ones aren't there yet.

@barbarabeeton Just ask M-x doctor. No AI problems and always so relevant answers :D

@barbarabeeton I am quite sure that there will always be people who value quality (which in this case would be non-AI generated code that has been looked over by a skilled human) ... I think I know a few already ... =)

@barbarabeeton or the general ones get better. as I say ask chatgpt a python question, the answer is better than its tex answer and (usually) it fires up an online python notebook and executes its code as confirmation, the fact that it hasn't seen as much tex is a symptom of the stage of development not of the technology in general

@DavidCarlisle -- Isn't that at least partly a function if the fact that TeX is relatively open ended, and even so, new features, or at least capabilities, keep being requested?

@barbarabeeton Well, all computer languages are relatively open-ended. Of necessity.

I would have thought something as idiosyncratic as TeX would be hard for a machine to get a handle on, but I could be wrong. Computers seem to be able to do many things that I would not expect them to be able to do. At least, not well.

@DavidCarlisle How does it handle hard questions? Say something where the answer isn't already on the internet in some form?

@FaheemMitha -- I'll be more amenable to persuasion when such a program can explain, in human-understandable terms, the workings of xii.tex, or xii-lat. Or, better yet, produce such a nice "brick" from a similar text in another language. Or even in English -- "Green grow the rashes" might be a nice starting point.

@TeXnician -- Are you one of the recipients of an early version of the writeup for my talk? Who was going to use the ideas in a tutorial?

@barbarabeeton No, I am not. I think it was @LaTeXereXeTaL who used your ideas in his LaTeX introduction.

@TeXnician -- Yes -- I'm sure you're right. (And your memory is clearly better than mine for such things.) Thank you.

@barbarabeeton The only thing about TUG talks I know that I have finished preparing mine and am waiting for @PauloCereda to start working on his. Otherwise I have no clue about TUG ;)

@TeXnician -- Well, in that case, be sure to look me up and introduce yourself. What's the subject of your talk?

@barbarabeeton I'm sure I will. I will be giving the Island of TeX talk about TeX Live and ConTeXt in a dockerized environment (tug.org/tug2023/abstracts/island-docker.txt).