www.digitalmars.com         C & C++   DMDScript  

digitalmars.D - Generative AI and D: an open letter

reply Indraj Gandham <newsgroups indraj.net> writes:
https://indraj.net/open-letter-to-dlf.pdf

This letter has now been sent to DLF. I would like to thank all 
of the signatories for their support.

Kind regards
Indraj
Jul 05
next sibling parent reply Dejan Lekic <dejan.lekic gmail.com> writes:
On Sunday, 5 July 2026 at 09:47:34 UTC, Indraj Gandham wrote:

 This letter has now been sent to DLF. I would like to thank all 
 of the signatories for their support.


Even though I do use LLMs in many places I would have signed this 
document myself (if i was asked) as I firmly believe D community 
has enough brains smarter than mine to maintain and improve 
current core D projects.

Sure I am not against core D devs using LLMs as "sidekicks" that 
can help them get some answers about codebase, review new changes 
before they are pushed to the official review, etc.
Jul 05
next sibling parent Indraj Gandham <newsgroups indraj.net> writes:
On Sunday, 5 July 2026 at 10:16:37 UTC, Dejan Lekic wrote:

 Even though I do use LLMs in many places I would have signed 
 this document myself (if i was asked) as I firmly believe D 
 community has enough brains smarter than mine to maintain and 
 improve current core D projects.


The letter is a live document, so you can still sign it and I 
will update it.


 Sure I am not against core D devs using LLMs as "sidekicks" 
 that can help them get some answers about codebase, review new 
 changes before they are pushed to the official review, etc.


I don't want this letter to be perceived as being anti-AI. It 
acknowledges that LLMs can enhance productivity, but at the same 
time highlights a very specific risk that members of our 
community are concerned about.
Jul 05
prev sibling parent Guillaume Piolat <first.name gmail.com> writes:
On Sunday, 5 July 2026 at 10:16:37 UTC, Dejan Lekic wrote:

 Sure I am not against core D devs using LLMs as "sidekicks" 
 that can help them get some answers about codebase, review new 
 changes before they are pushed to the official review, etc.


So far it seems to be the biggest AI risks is widespread "AI 
psychosis", the more you talk to AI, the more it creates 
psychosis and weird delusions. But D being good right now is some 
evidence from judgement not having been clouded by delusions in 
the past. So it would help if we are really sure not to "give up" 
and embrace magical thinking and psychosis.

Using LLMs for review (not letting them create) is on the other 
hand rather fruitful, like super good static analysis. Going 
full-vibe creates a sort of agitation that feels good but 
stressful a bit like hypomania.

That is to say, the letter highlight the copyright risk, but 
underneath there is imo the much uglier problem of destroying 
community.
Jul 05
prev sibling next sibling parent reply Sergey <kornburn yandex.ru> writes:
On Sunday, 5 July 2026 at 09:47:34 UTC, Indraj Gandham wrote:

 https://indraj.net/open-letter-to-dlf.pdf

 This letter has now been sent to DLF. I would like to thank all 
 of the signatories for their support.

 Kind regards
 Indraj


I think having relevant materials in one place is a good practice.
So I would ask everyone to post useful links here (other 
discussions inside D community or other language communities, 
lawyers recommendations, best practices from other companies and 
open source projects).

One small discussion was happened here for example: 
https://github.com/dlang/dmd/pull/22502
Jul 05
parent monkyyy <crazymonkyyy gmail.com> writes:
On Sunday, 5 July 2026 at 12:20:52 UTC, Sergey wrote:

 On Sunday, 5 July 2026 at 09:47:34 UTC, Indraj Gandham wrote:

 https://indraj.net/open-letter-to-dlf.pdf

 This letter has now been sent to DLF. I would like to thank 
 all of the signatories for their support.

 Kind regards
 Indraj


 I think having relevant materials in one place is a good 
 practice.
 So I would ask everyone to post useful links here (other 
 discussions inside D community or other language communities, 
 lawyers recommendations, best practices from other companies 
 and open source projects).

 One small discussion was happened here for example: 
 https://github.com/dlang/dmd/pull/22502


anti-ai wsj-writers: https://youtu.be/Pr6tOIjFXDs
cocaine-driven vibecoding: 
https://www.youtube.com/live/VNbW4wrcfSw
actual machine-learning content: 
https://www.youtube.com/watch?v=uQUsAndESdQ 
https://youtu.be/2eoOqqLvkl8
Jul 05
prev sibling next sibling parent reply Walter Bright <newshound2 digitalmars.com> writes:
LLM policies on other projects similar to ours, for comparison:

Rust:

Code creation: You cannot use AI (like Large Language Models) to write the code 
you submit.The human rule: You are completely responsible for your submissions. 
You cannot claim AI output as your own work.Allowed AI uses: You can use AI to 
hunt for bugs, analyze code, or review changes.Penalties: Low-effort, 
AI-generated code dumps (sometimes called "slop") can get you banned from
making 
contributions.

GNU:

GCC, and the GNU C Library) generally reject or severely restrict Artificial 
Intelligence (AI) and Large Language Model (LLM) generated code

LLVM:

Human Vetting: You must read and understand all AI-generated code before asking 
project members to review it.Labeling: You should label contributions that 
contain substantial amounts of AI-generated content.No Autonomous Agents: 
Digital bots or agents that submit code or make changes without human approval 
are not allowed.No AI for Beginners' Tasks: AI tools must not be used to solve 
issues labeled as "good first issues," as these are meant to be learning 
opportunities for new contributors.

Python:

The Python Software Foundation (PSF) and the core Python language place zero 
restrictions on how you write your code.

Java:

The OpenJDK Governing Board officially bans AI-generated code from being 
submitted to the project. You can use AI as an assistant to brainstorm, but you 
cannot submit code written by a Large Language Model (LLM) into the official 
Java codebase.

Zig:

Zig has a strict ban on AI-generated code.The Zig language has an official rule 
against submitting any LLM-generated content. This includes brainstorming, 
translating, debugging, or writing code.
Jul 06
next sibling parent reply Indraj Gandham <newsgroups indraj.net> writes:
Some of the policies you cite contain provisions related to code 
quality. This is not the issue highlighted in the letter.

It has become known that certain modules which contain large 
amounts of LLM-generated code are being considered for inclusion 
in Phobos v3. The signatories of the letter oppose this course of 
action on the grounds that such code is of questionable 
provenance and request that DLF seeks legal advice before doing 
so.

I think this is entirely reasonable since there is at least one 
ongoing lawsuit in the United States which specifically 
references model output as being potentially infringing. We 
probably won't know the final outcome for a while as the parties 
involved are likely to appeal any rulings all the way to the 
Supreme Court.

Phobos is in a unique position compared to other D components as 
portions of it are frequently included in or distributed with 
application executables. Since it is the official standard 
library, it is expected that most applications will make use of 
some of its functionality.

In response to a user suggestion in another thread that Phobos 
should segment LLM-generated code, it was made clear by one of 
the developers working on v3 that "all of the stdlib would end up 
in the LLM segment".

This is a highly complex legal matter and it requires an answer 
from a legal professional. Many of those projects you mentioned 
have large corporate sponsors and have nothing to worry about 
even if their policies turn out to be legally insufficient, as 
these sponsors will be able to allocate the resources required to 
rectify any mistakes.

As far as I am aware, D does not enjoy the same degree of 
support. It's a bit like gambling in Vegas when you're broke just 
because all your friends are doing it.

D is going to have to find its own path, but seeking and 
publishing advice from someone qualified to provide it would be a 
good first step towards building confidence in any policy DLF 
wishes to adopt.
Jul 06
next sibling parent reply Adam Wilson <flyboynw gmail.com> writes:
On Monday, 6 July 2026 at 23:27:58 UTC, Indraj Gandham wrote:

 In response to a user suggestion in another thread that Phobos 
 should segment LLM-generated code, it was made clear by one of 
 the developers working on v3 that "all of the stdlib would end 
 up in the LLM segment".


You are taking my comment entirely out of context and deceptively 
framing it. The users suggestion was to preemptively segment 
*ALL* code that was touched by LLM's in *any way*. My response 
was that over time, as LLM's touch more and more of the code, 
then in short order "all of the stdlib would end up in the LLM 
segment". Which is objectively correct.

You are trying to make it sound like I said that we would be 
forcing all Pv3 code into the LLM segment specifically so that we 
can freely use LLMs on Phobos 3. This is false. The truth of my 
statement is self-evident in it's full context and is far less 
provocative than your framing of the quote would suggest. I was 
describing the end-state of the users request. It was *NOT* a 
position statement about how Phobos 3 development is to be 
conducted.

If you are going to make an argument using my words, please have 
the basic courtesy to include the full context. You are not doing 
your argument any favors by cherry-picking a quotes then using a 
deceptive framing on it to make it sound like I said something 
far more provocative than I did. You will not win debates by 
deceiving your audience.

(Link to the referenced post for any interested readers: 
https://forum.dlang.org/post/vekjnpgkwkspznqpceej forum.dlang.org)
Jul 06
parent Indraj Gandham <newsgroups indraj.net> writes:
The author of that post did not explicitly specify the degree of 
LLM use. However, he did say:


 If the upstream devs is going to start using ai, this should be 
 preemptively segmented into a new namespace, perhaps `bracca`. 
 Rather then trying to clean up a mess later, in the event 
 copyright law suddenly declares you must back pay if you used 
 ai trained on gnu code.


I don't think this could be a problem if the LLM was being used 
only for small edits. You have assumed meanings and intentions in 
both that post and my post here. The author also clarified in his 
reply to you that he was not suggesting that LLMs should be 
banned, as you assumed.


 You are trying to make it sound like I said that we would be 
 forcing all Pv3 code into the LLM segment specifically so that 
 we can freely use LLMs on Phobos 3. This is false. The truth of 
 my statement is self-evident in it's full context and is far 
 less provocative than your framing of the quote would suggest. 
 I was describing the end-state of the users request. It was 
 *NOT* a position statement about how Phobos 3 development is to 
 be conducted.


I was not insinuating this at all. I brought it up as it suggests 
that the segmentation approach has already been rejected.


 If you are going to make an argument using my words, please 
 have the basic courtesy to include the full context. You are 
 not doing your argument any favors by cherry-picking a quotes 
 then using a deceptive framing on it to make it sound like I 
 said something far more provocative than I did. You will not 
 win debates by deceiving your audience.


It would seem that you are not particularly interested in 
debating this issue anyway:

https://forum.dlang.org/post/adajbcujigtteasbpayk forum.dlang.org

But I suppose I shouldn't assume.
Jul 06
prev sibling next sibling parent Walter Bright <newshound2 digitalmars.com> writes:
Thank you for the considered and thoughtful reply.
Jul 06
prev sibling parent Denis F <denis.feklushkin gmail.com> writes:
On Monday, 6 July 2026 at 23:27:58 UTC, Indraj Gandham wrote:


 I think this is entirely reasonable since there is at least one 
 ongoing lawsuit in the United States which specifically 
 references model output as being potentially infringing. We 
 probably won't know the final outcome for a while as the 
 parties involved are likely to appeal any rulings all the way 
 to the Supreme Court.


Most of us definitely use copyrighted sources to commit code to 
the D language repository. At least, my own natural neural 
network is trained on texts from books the rights to which belong 
to various publishers.
Jul 06
prev sibling parent Denis F <denis.feklushkin gmail.com> writes:
On Monday, 6 July 2026 at 22:12:28 UTC, Walter Bright wrote:


 Zig:

 Zig has a strict ban on AI-generated code.The Zig language has 
 an official rule against submitting any LLM-generated content. 
 This includes brainstorming, translating, debugging, or writing 
 code.


A rule is nonsense without the ability to enforce it. At some 
point means to distinguish generated code from human-written code 
will disappear entirely.

Why are we even worried? Everything is fine as long as human eyes 
are reviewing PRs.
Jul 06
prev sibling parent user1234 <user1234 12.de> writes:
On Sunday, 5 July 2026 at 09:47:34 UTC, Indraj Gandham wrote:

 https://indraj.net/open-letter-to-dlf.pdf

 This letter has now been sent to DLF. I would like to thank all 
 of the signatories for their support.

 Kind regards
 Indraj


I cant sign. Your concern is about licences. My concern is more 
about the loss of knowledge. Globally however I agree that people 
should take care.
Jul 07