[tldoc] TeXLive plans in writing the documentation: will be human or chatgpt generated text?

[Carlos]

On Thu, Jun 08, 2023 at 03:00:21PM -0700, Boris Veytsman wrote:
> KB> Date: Thu, 8 Jun 2023 15:01:26 -0600
> KB> From: Karl Berry <karl at freefriends.org>
> 
> KB> Carlos - TUG as an organization has no opinion pro or con for chatgpt,
> KB> or llm in general. The board has never discussed it; as far as I can
> KB> see, there's nothing to discuss. In his newsletter, Boris was simply
> KB> making an observation. Boris is on this list and can confirm if it would
> KB> ease your mind.
> 
> 
> Absolutely.
> 
> It is true that LLMs now are a hot topic, so the ability of them to
> produce TeX code (or rather LaTeX code (*)) is rather amusing.  Still the
> amount of confabulation by them is too large (I think the technical
> term is hallucination).  I reported elsewhere the abysmal results of
> them for a problem which computers have been able to solve for fifty
> years (https://tug.org/TUGboat/tb44-1/tb136pres.pdf).  There are
> horror stories about LLMs inventing precedents for legal briefs with
> rather bad consequences for the lawyers.

Interesting Boris.

> 
> As far as I know there are no plans of using LLMs in documentation
> writing or code generation for TUG.  If these projects appear, it
> would be interesting to see the results - and we cannot exclude the
> possibility they might be useful.  Especially if we can solve the
> hallucination problem.

What results specifically would you expect to find? Karl wrote earlier:

> I can tell you that none of the projects I am involved, including the
> various manuals, are currently using or plan to use chatgpt or any other
> llm.  I mean, if a translator wishes to use chatgpt to help, that's
> fine, but it's up to them. I neither object nor suggest. Nor do I have
> any way of knowing if someone does.

I'm not a lawyer. And needlessly to say, my level of knowledge in this
area is archaic, rudimentary. But if any of these LLM programs (chatgpt and
its ilk) go well beyond a mere grammar substitution that seems to
be currently accepted by expounding the differences in code as you
posited below among plain TeX and (La)TeX, what is there to stop it
from using just more sophisticated methods and change definitions
along with the code therein, and while at it, change the licenses too?

The latter, it follows, would even be easier once the remainding chunk is
modified.

How will the lawyers from a specific region set precedents in other regions? 

Will the lawyers from a specific region debrief their arguments by
generating them with LLM programs too? Will it be legally permissible
to be assisted by LLM programs while debriefing?

How will the laws be drafted in other regions? By using LLM programs?

How would humans be able to spot the difference/s? 


> 
> ---------------------------------------------------------------------------
> (*) Another amusing thing - it looks like chatgpt understands the
> difference between LaTeX and plain TeX:
> 
> BV: hi
> 
> ChatGPT: Hello! How can I assist you today?
> 
> BV: can you translate LaTeX code to plain TeX?
> 
> ChatGPT: Certainly! I can help you with that. Please provide the LaTeX
> code you'd like to translate, and I'll do my best to convert it to
> plain TeX for you.
> 
> BV: please translate $\frac{a}{b}$ to plain TeX
> 
> ChatGPT: Certainly! In plain TeX, the equivalent code for the fraction
> \frac{a}{b} in LaTeX is: $$\displaystyle{a \over b}$$ This will
> produce the same fraction a/b in plain TeX.
> 
> 
> -- 
> Good luck

After reading the prior exchange, allow me please to just type a colon
and a closing parenthesis as pictograph/characters forms would do,
resembling, most likely, how the same would be closely described by
an LLM program

:)

> 
> -Boris
> 

-- 
The first version always gets thrown away.