Re: reading LLM code may rule out white room re-implementation; was: On keybindings and the slow erosion of help's utility

On 2026-07-18 07:46, Dr. Arne Babenhauserheide wrote:
> Jean Louis <[email protected]> writes:
> 
>> On 2026-07-17 18:57, Dr. Arne Babenhauserheide wrote:
>>>> "Dr. Arne Babenhauserheide" <[email protected]> writes:
>>>>> So if code under an incompatible license may prove to have been
>>>>> material
>>>>> to the LLM output, and if the legal landscape changes so LLM
>>>>> processing
>>>>> of code preserves some copyright (which currently seems likely), 
>>>>> then
>>>>> LLM code posted here may make it illegal for anyone who read it to
>>>>> write
>>>>> a free software implementation of the idea.
>> 
>> Those are such extreme ideas yet not well researched
> 
> There are nuances, read up yourself:
> https://en.wikipedia.org/wiki/Clean-room_design

Sorry, your generalized statement is not supported by Wikipedia article 
you have sent to me.

All I am saying, please be specific among millions of models, so name 
the model for which you think it applies.

You cannot possibly make general statements and put people into fear.

How do you even know which LLM I am using on my computer? Do you know 
the model name, version? Who made it? How to claim that each LLM in the 
world could output code and make it illegal here for anyone who read to 
write free software implementation of the idea?

You cannot by anything even prove it.

All that it brings is FUD.

How about you read this one:

Apertus | Swiss AI:
https://www.swiss-ai.org/apertus

Does your statement apply also to that model? Could you provide a source 
of copyrighted materials from that model?

Be scientific. Show some examples.

>>> For LLM code you currently can’t know whether you will own it a year
>>> from now.
>> 
>> Inaccurate. Not true! Reasons explained above. And yes, it is quite
>> possible to know whether LLM code generated one will own it or not.
> 
> Do read:
> https://en.wikipedia.org/wiki/Artificial_intelligence_and_copyright#Training_AI_with_copyrighted_data
> 
> And please don’t pile walls of text upon me. Five minutes of searching
> (not even research) would have sufficed.

I am definitely not writing for you yet for readers to get warned that 
generalized statements you provided simply do not match in many cases. 
Like in the case of Apertus model, do you really wish to claim your 
statement is true?

How about other millions of models. You have so far missed to mention 
model names, so saying "For LLM code you currently can't know whether 
you will own it a year from now" -- that is so generalized.

Did you ever try or attempt to make even the smallest language model 
yourself? I suggest doing that exercise.

Would you then in that case of training your own language model know if 
it was trained on copyrighted information? So ask yourself how 
generalized statement impact other people who do take care of which 
model they are using.

Please not accuse random people and don't spread FUD.

-- 
Jean Louis