[Libre-soc-dev] [netsurf-dev] RISCsurf patches

[lkcl]

(please note that i have PTSD and have had
multiple strokes and heart attacks due to the
complex interaction of prior-unknown life-long
medical disabilities, Domestic Abuse and more.
please respond with respect and care
and *under no circumstances* engage in
"behind my back" discussion as has previously
occurred. that is specifically directed to Vince,
from whom i am still waiting on for an apology)

On Tuesday, October 27, 2026, John-Mark Bell <dmarc-noreply at freelists.org>
wrote:

Additionally, the use of LLMs does not make you any less responsible for
> the contents of submitted patches. You must, for example, still be able to
> claim ownership of the work to be able to make it available for inclusion
> in NetSurf under our licensing terms (i.e. GPLv2 for the browser itself and
> other licenses for certain other components).


so i have a unique perspective on this. strokes
*forced* me to begin using AI (note that i did not
state LLMs, more on that in a bit) - to compensate
for the damage that lost me the ability to perform
basic math.

the damage occurred to the area that covers a link
between words and math. i had no idea what that
entailed, and very distressingly quickly found out
(and got zero help or support. had to do my own
research)

i lost verbal-math times tables i learned aged 9, i
lost power-of-two arithmetic capacity since age 14,
and hexadecimal conversion from binary.

as you can probably understand this was terrifying
and distressing to find it took 5 seconds per
number to verbally internally count 2 4 8 10 12 16
only to find that you missed out 6 and 14, and have
a severe headache from the effort.

i therefore started investigating commercial AI
on sponsored subscription and after many months
of training the AI (fighting OpenAI's horrendous
inept default instructions) found that my Software
Engineering expertise could make me over ONE
HUNDRED times more effective at writing software
[caveat: *if* we had the right discussion in advance]

50,000 lines of code including unit tests could be
written in TWO DAYS. supporting documentation of
up to 150k in length was not in any way a problem.

caveat: when the AI is not properly set to the
correct "high" level however, such that it fails
to understand your instructions, it all goes to shit
very VERY quickly.

FOSS volunteers - many of whom object very
reasonably to the cost, the lack of attribution,
and to the forced provision of information to
Corporations that will not have their best interests
at heart, simply cannot keep.

i am... humiliated at having no choice due to the
strokes but to abandon the ethics and principles
i've held for decades, and have set myself the goal
of bootstrapping up a proper FOSS AI ironically
*using* commercial AI to create it, via complex
firebreaks based on reverse-engineering experience.

long story, but important in support of John's warnings.

John: it is actually much worse than you suspected
or expected, but first: a Large Language Model is
a static database dump. it's not a library or a program that uses reads *or
writes* in the data
*format* behind an LLM.

saying "i use LLM" is like saying "i used a psql-dump
to create a stonking powerful Geodata website"
when you meant to say you wrote a *program*
that used the postgresql *server* with the postgis
*plugin* using SQL *queries* on SQL *data* to
create a powerful GIS website.

AIs *use* LLM data *formats* with a library that
allows dynamic read-write in a very complex
hierarchical OO form that is able to do way more
than any 4th Normalised Form SQL database ever
could.

and OpenAI for example managed to severely
damage their AI's capability by *changing that
library* without performing months of re-training,
resulting in the AI subtly going down *untested*
LLM data retrieval paths and consequently corrupting
answers, retrieving words that are completely
the wrong synonym (especially on foreign languages)
or in worst case being unable to complete tasks it
was previously successful at!

to the reasons why i say it is much worse than you
might think:

1. the training on learning a given programming
    language has been done on FOSS code where
    attribution has not been preserved.

    you have NO IDEA if the lines of code being
    given to you as output from an AI actually
    originated from another FOSS project, let
    alone know what the License is. was.

    worse: the AI *itself* can and will go out on
    the internet, even if it has been explicitly
    default-firewalled, hunting for answers and
    potentially "creating" code WITHOUT YOUR
    KNOWLEDGE that is literally lifted from a
    website where you AS ITS USER forgot to
    tell the AI "if you find some code please tell
    me where you got it from", and even if you did
    tell it IT FORGOT TO CORRECTLY FOLLOW YOUR
    EXPLICIT INSTRUCTIONS

    this latter problem is getting much worse as i
    have a sneaking suspicion that recent political
    demands to "rein in" AI may have resulted in
    deeply significant and subtle damage to their
    *capacity to understand*.

     (several people have independently noticed
      adverse behavioural changes to both GPT and
      Claude in the past few weeks)

2. a copyright check - such as that required by
     Debian - therefore effectively becomes a
     false declaration if the origin of the copyrighted
     material is smeared holographically across
     a *proprietary* Corporation's databases then
     *claimed* to be an individual's copyrighted
     material when actually they used a *tool*
     that snaffled the source from unattributed
     Copyrighted sources and *you can't check*.

     git commit -s with a cryptographic identity
     check would at least help you move the
     responsibility, legally. but you likely know
     full well the lengths that the debian project
     go to on GPG keyring signing [passport
     or other ID checks *in person only*]

3. Anthropic has had to settle a massive class
     action lawsuit with Academic Journals, every
     author qty millions receiving $3,000 after it was
     discovered that peer-reviewed text from
     published academic work was "turning up"
     in AI-assisted scientific text WITHOUT
     the usual Academic attribution / citations.

     *unknowing* plagiarism => Class Action Lawsuit.

4. FOSS absolutely does not have that kind of
     financial legal clout and consequently is getting
     effed over preeeetty much in exactly the same
     way that my work has been systematically
     abusively exploited for decades with no credit
     or financial reward... except this is WHOLESALE.

     [in all seriousness i take no pleasure from being
     a witness to that. i can't. i won't go into the
     medical details of the toxic effects of the
     "usual" neurochemicals behind hate and revenge,
     but whilst most people recover from having such
     toxic "feelings", for me the downstream effects
     can literally place me at risk of death from a
     heart attack or stroke.]

5. the GNU Project is assessing quotes ell ell emmz
     quotes (sigh) and has requested that its
     developers firstly cease using AIs and secondly
     cease accepting patches from quotes ell ell
     emmz sigh quotes until they have had
     a Legal Opinion provided and have considered
     the implications very very carefully.

i think that's probably enough. regarding this
particular example can i strongly recommend
looking at Trademark Registration of netsurf
as it offers a degree of protection that Copyright
simply cannot. your *service* is to develop the
netsurf browser, and its replication and insertion
of potentially unattributed Copyrighted material
could get *you* into legal trouble in some
jurisdictions if *you* get blamed for bugs
introduced by an inadequately-correctly-instructed
AI.

https://www.netsurf-browser.org/about/licence.html?utm_source=chatgpt.com

no Trademark registered there *but you still have one*,
it's termed an "Unregistered" Trademark.

IANAL but my *understanding* is that Trademark
Law will help you take legal action against an
irresponsible fork where Copyright Law just...
won't

and reminder: the basis of Trademarks is the
potential for confusion in the minds of users.

l.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.libre-soc.org/pipermail/libre-soc-dev/attachments/20261028/345f9857/attachment-0001.html>