Re: GPT-4 knows Guile! :)

[Zelphir Kaltstahl <zelphirkaltstahl@posteo.de>]

On 3/20/23 00:18, Dr. Arne Babenhauserheide wrote:
> <tomas@tuxteam.de> writes:
>> which has tried to be free (in both senses). For images, there is
>> Stable Diffusion (disclaimer:I don't know much about them). For
>> raw data, there's Common Crawl [3], which is Stable Diffusion's
> Stable diffusion has one of the most evil licenses I’ve read till now:
> using its images seems to be allowed for almost all proprietary uses,
> but illegal for free culture use. See the explicit questions about that
> here: https://github.com/CompVis/stable-diffusion/issues/131
>
> A license that makes its output only usable for proprietary creations
> but not for free culture is like the polar opposite of the GPL or CC
> by-sa: making a tool that gives proprietary creations an advantage over
> Free Culture.

That would double evil, since they most likely trained their stuff on freely 
available things. Basically taking from our culture, but not giving anything 
back and netting the profits.

My biggest concerns about these big machine learning models are:

(1) The laundering character, be it code, other texts, prose and to on, or 
artworks.—What and where are their sources? Complete list of inputs? Where is a 
form, that allows me to make my works invisible to past, present and future 
models they produce, without being tracked by Google and its ilk? How will 
artists and others be able to safely, securely, and unambiguously tell them 
"no"? Oh wait, it is basically impossible to remove something from a learned 
model after the fact. Duh! How about asking beforehand?

In the end it is the same typical kind of shitty behavior we see from entities 
like FB, oh wait, they call themselves "Meta" now lol.: "Oh what you want is 
soooo complicated to do and would be sooo much effort! We cannot possibly be 
expected to do that, so sorry!"—Well then don't start doing what you do in the 
first place! If doing it ethically does not scale, then don't friggin start scaling!

(2) Establishment of paywalls and unethical registration forms. Soon some 
employers might ask their employees to "be more efficient" by using GPT models 
and either buying licenses, or even demanding employees buy them themselves, 
while of course completely neglecting to think about the license terms or what 
happens to the data we input into those prompts. Registration might even require 
us to input personal data, like where we live or our names.

(3) Also no one seems to think about not being able to use these tools on 
ethical grounds: I visited the GTP website and thought: "OK lets give it a 
try.", but then I could not even do that, unless I allowed tech giants to track 
me. No thanks! Tab closed. Given how they seems to act so far, I do not expect 
them to set up any sane login or website any time soon.

(4) It might take years again before any of that gets properly regulated. First 
it might be shittly regulated, because of governments not asking experts or if 
asking experts asking the ones with heavy bias for their employers (like 
consulting Microsoft for publicly in government used software …).

Regards,
Zelphir

-- 
repositories: https://notabug.org/ZelphirKaltstahl