Bitcoin Development Mailinglist
 help / color / mirror / Atom feed
From: Jon Atack <jonnyatack@gmail.com>
To: Bitcoin Development Mailing List <bitcoindev@googlegroups.com>
Subject: Re: [bitcoindev] Motion to Activate BIP 3
Date: Wed, 19 Nov 2025 13:21:00 -0800 (PST)	[thread overview]
Message-ID: <624c9e1f-ffb8-4165-83f4-d445f8777476n@googlegroups.com> (raw)
In-Reply-To: <CABaSBaxR0ytoHHFEAe-+49VSWPaYan6N3J1DdG-51LK-d49xWA@mail.gmail.com>


[-- Attachment #1.1: Type: text/plain, Size: 7776 bytes --]

Hi list,

I agree with Greg Maxwell's thoughts on AI use.

We have been seeing quite a bit of AI-generated PRs being thrown over the 
fence at us in the BIPs repository.

In a few cases, the proposed fixes are useful.

In many others, they seem to be a waste of review/maintenance/moderation 
bandwidth and time, and are demotivating to deal with.

I am also aware of BIPs that turned out to have been largely generated or 
rewritten to final form by LLMs. I am not a lawyer, but the copyright 
situation WRT such BIPs seems unclear and evolving. This was not the 
principal motivation of the BIP 3 addition about AI/LLM use -- not wasting 
our time is -- but is a context that may need keeping tabs on.

Personally, I don't wish to spend time manually reviewing AI slop or 
hallucinations.

Due to what we're seeing and to what Greg rightly describes above, I think 
we'll need to set up auto-review by LLMs in the CI pipeline for (a) a 
degree of confidence grade of AI being used, and (b) perhaps an initial 
technical review. Similar to Drahtbot in Bitcoin Core and to what I see in 
use in other organizations that I review for.

For now, these tools will vitally need to be followed by human review and 
subjective judgment calls on our part.

The BIP 3 section about reducing judgment calls is probably unrealistic at 
this time and in need of an update before activation.

Best.

On Tue, Nov 18, 2025, 7:18 PM Greg Maxwell <gmax...@gmail.com> wrote:

No doubt *you* are able to make good documents with or without the aid of 
AI.

With outright AI 'authorship' you immediately run into potential copyright 
issues-- which I think is the origin of the "generated by" prohibition, 
otherwise I think disclosure would be sufficient.

Taking a step back: is Bitcoin's welfare maximized by permitting LLM glurge 
submissions in standards documents? In some cases it's benign, I readily 
agree, in others its harmful.  But the number of good submissions that 
could be made would hardly be increased by LLMs (being limited by expert 
proposers with good ideas) but the number of potential poor submissions is 
increased astronomically.  So I think it's pretty clearly a net harm to 
have text authored that way.

I've never had an impression that drafting was at all a limiting step in 
writing BIPs, though even to the extent that it has been at times it's 
possible to use LLMs in a review capacity to make authorship much easier 
("What's missing / unclear?") without resorting to using it to author.

There is a particularly clear pattern at least with current LLM tools that 
users who lack the skills to have authored the work without an LLM are 
generally unable to recognize when the LLM is full of crap (and even 
sometimes when they should know better), so unfortunately they're only 
benign to use in the hands of those whose need is the least.  

And as a reviewer outside of Bitcoin I've found LLM powered proposers to be 
absolutely the worst to deal with. Because they're not submitting their own 
words and ideas, they're unable to change their thinking in response or 
explain sufficiently to change yours--- the interactions often degrade to 
them just copy and pasting their chatbot back to you.  Because it's cheap 
to generate more text they also tend to flood you out with documents 
several times longer than any human author would have bothered with.

I think LLMs have generally created something of an existential threat to 
most open collaborations: Now its so easy to get flooded out by subtly 
worthless material.  Many projects, including, Bitcoin have long struggled 
with review capacity being limited and a far amount of time waste by 
thoughtless (or even crazy!) submissions, but now it's automated and even 
the most well meaning person may now make submissions that are as bad as 
the most deviously constructed malicious submissions could have been in the 
past, not even know they are doing it, and can make a dozen proposals 
before lunch without even breaking a sweat.



On Wed, Nov 19, 2025 at 12:06 AM David A. Harding <da...@dtrt.org> wrote:

On 2025-11-04 15:10, Murch wrote:
> Summary of changes since BIP 3 was advanced to Proposed:
> [...]
>   - that BIPs submissions may not be generated by AI/LLM⁵
> [...]
> ⁵ https://github.com/bitcoin/bips/pull/2006

I strongly disagree with this change.  If I were to begin working on a 
new BIP today, I would use AI throughout the process.  I'd ask it to 
help me create a todo list of what should go in the BIP; I'd ask it to 
create a draft based on existing BIPs, my todo list, and whatever other 
work products I had (e.g. prototypes); I'd then ask it to help me refine 
the document until I was satisfied.

I would, of course, review every word of the draft BIP before submitting 
it for consideration and ensure that it represented the highest quality 
work I was able to produce---but the ultimate work would be a mix of AI 
and human writing and editing.

I think considerate use of AI would be even more valuable for people who 
are less comfortable with writing technical English-language documents 
than I am.  For example, non-native literates, people with disabilities 
that make text input difficulty, and those who recognize that they're 
bad writers.

The PR forbidding AI doesn't go into any detail about its motivation, 
although it references a previous discussion[1] where a low-quality BIP 
PR was opened using mostly AI-generated content.  I'm guessing the 
motivation is that AI (by itself) generates low-quality technical 
content, BIPs should be high-quality technical content, and therefore we 
should ban the use of AI.

However, as mentioned in the previous discussion, the BIP process 
already requires high-quality content.[2]  AI-generated content can be 
high-quality, especially if its creation and editing was guided by a 
knowledgeable human.  Banning specific tools like AI seems redundant and 
penalizes people who either need those tools or who can use them 
effectively.

I advocate for reverting the first hunk of BIPs repository PR 2006.

-Dave

[1] https://github.com/bitcoin/bips/pull/2005
[2] "After fleshing out the proposal further and ensuring that it is of 
**high quality** and properly formatted, the authors should open a pull 
request to the BIPs repository." --BIP3, emphasis added

-- 
You received this message because you are subscribed to the Google Groups 
"Bitcoin Development Mailing List" group.
To unsubscribe from this group and stop receiving emails from it, send an 
email to bitcoindev+...@googlegroups.com.
To view this discussion visit 
https://groups.google.com/d/msgid/bitcoindev/3a66dbbe9a9c46566c8a9a16ccb1cc91%40dtrt.org
.

-- 
You received this message because you are subscribed to the Google Groups 
"Bitcoin Development Mailing List" group.
To unsubscribe from this group and stop receiving emails from it, send an 
email to bitcoindev+...@googlegroups.com.

To view this discussion visit 
https://groups.google.com/d/msgid/bitcoindev/CAAS2fgRV1aZ9xvAhBriZ%3DXdmYf5CvrvXWXsjVD07uynivW_qkg%40mail.gmail.com 
<https://groups.google.com/d/msgid/bitcoindev/CAAS2fgRV1aZ9xvAhBriZ%3DXdmYf5CvrvXWXsjVD07uynivW_qkg%40mail.gmail.com?utm_medium=email&utm_source=footer>
.

-- 
You received this message because you are subscribed to the Google Groups "Bitcoin Development Mailing List" group.
To unsubscribe from this group and stop receiving emails from it, send an email to bitcoindev+unsubscribe@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/bitcoindev/624c9e1f-ffb8-4165-83f4-d445f8777476n%40googlegroups.com.

[-- Attachment #1.2: Type: text/html, Size: 9983 bytes --]

  reply	other threads:[~2025-11-19 21:46 UTC|newest]

Thread overview: 31+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2025-11-05  1:10 Murch
2025-11-05  1:53 ` Ruben Somsen
2025-11-12 19:03 ` [bitcoindev] " Greg Sanders
2025-11-13  0:23   ` Murch
2025-11-13 18:54     ` Greg Sanders
     [not found] ` <Qe_CRlsuwalN-0a0oo1KcZ265rXevTeHaTdk6IifH-j7NbLMew7-ucLLMiwECQLZEPoU2pm-PAuwb_lZAeCU9vChaVYTZzl60N9jyPTnUbo=@protonmail.com>
2025-11-13 19:35   ` [bitcoindev] " 'Antoine Poinsot' via Bitcoin Development Mailing List
2025-11-13 21:43 ` Murch
2025-11-14 17:05 ` Melvin Carvalho
2025-11-22 23:46   ` Murch
2025-11-29 23:00     ` Melvin Carvalho
2025-12-02  0:29       ` Murch
2025-11-15 22:01 ` Luke Dashjr
2025-11-18  4:26   ` Greg Maxwell
2025-12-02 22:46   ` Murch
2025-11-18 15:47 ` David A. Harding
2025-11-19  1:12   ` Greg Maxwell
2025-11-19  1:20     ` Bryan Bishop
2025-11-19 21:21       ` Jon Atack [this message]
2025-11-19  6:25     ` David A. Harding
2025-11-19  6:58     ` Bitcoin Error Log
2025-11-20  1:22       ` Bitcoin Mechanic
2025-11-20  9:06         ` Oghenovo Usiwoma
2025-11-20 17:23           ` Greg Maxwell
2025-11-22 15:14             ` Jon Atack
2025-11-22 19:30               ` Sjors Provoost
2025-11-22 20:53                 ` /dev /fd0
2025-11-28 15:35                   ` Tim Ruffing
2025-11-28 16:34                     ` Jon Atack
2025-11-28 19:21                       ` Pieter Wuille
2025-11-22 21:06                 ` Bill MacDonald
2025-11-20 20:14           ` Mat Balez

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=624c9e1f-ffb8-4165-83f4-d445f8777476n@googlegroups.com \
    --to=jonnyatack@gmail.com \
    --cc=bitcoindev@googlegroups.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox