I miss the old times, back when humans used to write Purism’s blog posts, human-style.
I dislike this new LLM-generated/-optimized breed of copywriting so much that by now, I feel completely unable to immerse into the content even a bit.
I cannot confirm whether AI was used as a primary source of authorship for this article, but the content does appear at least partially AI-generated based on its composition/styling. I hear your concerns and have edited Purism’s internal content creation guidelines to state that their content is human-first and not AI-first - that AI may be used to assist in proofreading, but should not be relied upon heavily for composition.
If you want human-first content, credit should be provided to authors:
Staff accounts are eventually closed per policy for security reasons. Kyle was a prior staff member with Purism and the IP remains with Purism. Kyle was not disattributed for any malicious purpose; Wordpress simply inherits the Purism author as a fallback. There may be opportunities for reattribution in the future as time and human resources allow.
My proposal is that accounts are not deleted/disabled but rather have write access revoked. I don’t have time to address this right now, but I can ensure there’s a ticket for tracking.
And this explanation was covered off back in September when @FranklyFlawless first raised the issue (topic already linked by Frankly).
I can only say what I said back in September … that is not security best practice. Having accounts still around that are not supposed to be being used at all but can still be used will not please your “security auditors”.
I guess a question that I would raise is … is your Content Management System capable of caching the author details - so that while the author account still exists, current details from the account will be looked up and once the account is deleted, cached details will be used?
An alternative might be a script that is manually run within the CMS immediately prior to account deletion that copies the current details from the account into the body of the authored document - rather than have someone do this manually - but in this particular case it is too late for this paragraph to apply.
This proposal was my best attempt at a compromise since this concern has been raised several times, but I understand if it may incur undue risk (and may possibly be the explanation of why this hasn’t been done thus far.)
It does not have this capability, from what I’ve seen.
Thank you JCS. It feels so good to be heard!
Is this actually true that Purism is writing this article with AI or is that forum slander? If so, what AI was used to create it? Are we using Facebook’s LLAMA AI to generate slander about Facebook or whatever? I mean I guess that’s funny on the one hand, but on the other hand, it would seem extremely disingenuous to expect already paranoid Purism hardware users (real humans) to read AI generated fluff about how bad Facebook is while generating that fluff with a Facebook technology. Just saying.
The claim does not have explicit evidence to either prove or disprove it.
You’re correct. I don’t have hard evidence. But I’ve seen my share of LLM-generated (or -overhauled) content, and I think I’ve grown overly sensitive for the tell-tale signs.
Regarding Purism’s post, I do have some circumstantial evidence. Each piece of evidence says nothing per se, but the combination is what made me confident enough to make that claim. Because it is so tell-tale that it killed my immersion.
-
Gratuituous usage of the em-dash, which is really, really uncommon in human-written language.
-
Humans occasionally use title case, especially in titles. Nothing wrong with that.
What LLMs do though is use goddamn title case in every single list item. It’s always the same pattern (emphasis added by me):
- Invasive Tracking: Meta collects extensive user data, often beyond what users explicitly provide.
- Algorithmic Manipulation: AI-driven content curation shapes what users see, reinforcing engagement loops that maximize advertising revenue.
- Lack of True Consent: While privacy settings exist, they are often buried in complex menus designed to deter users from opting out.
-
There’s not a single stylistic, grammatical, or typographic flaw in the whole text (except for the Digital Bill of Rights, see next point.) It’s written in perfectly crafted language. Extremely unusual and out of character for a small hardware company. Yes, genius writers exist who moonlight as niche phone manufacturers. But the contrast to earlier posts is so massive that it’s a bit hard to believe. Nothing wrong with letting AI proofread stuff. But it comes at a price, which is that readers might start asking questions.
-
The Digital Bill of Rights is rather messed up. It says 1, (a), 2, (b), and so on. As if someone (or something) that didn’t know the exact Markdown flavor of Purism’s CMS emitted some Markdown, which someone then went on to copy into the CMS, but the person who copied it didn’t look at the preview closely enough before they hit Submit. It doesn’t feel convincing to me why any human would craft an article with such linguistic perfection and then let such a blatant Markdown blunder pass.
I could go on. But there’s no point. It’s not proof, just a pile of circumstantial evidence. I might be completely misguided. But it was enough for me to bother complaining. Thanks again @JCS for taking my feedback seriously.
Thanks @Lliure , I appreciate the provided evidence.
I will admit, at some point recently on a very niche video game related website, I was in a situation where one guy asked another guy where his profile picture came from. I interjected between them and said, “it was generated by AI” because it portrayed a stylized cartoon of an attractive female. They were surprised I could tell, and then asked me for evidence, but in my case I didn’t have evidence. I was moreso accusing anything I encounter of being AI as I get more and more tired of it, and thought that in this case it was funny to do so.
But it’s quite interesting to think, suppose that I told them I was confident without providing any evidence. Can someone prove a negative? Could one of them prove it’s not AI, and prove me wrong? How would they do that? I can’t think of a way.
A good start would be providing credit to authors, but upholding that standard is considered too difficult for Purism to achieve now.
I looked at an article about the Apple trial a week or so ago. And maybe I am not able to see what is right and what is wrong, but it seemed like a bad direction for society. They were making some rule about the size and kind of redirect window Apple could provide to users when implementing the hyperlink function from in apps or app descriptions.
And I think it seems really important for society to have some common understanding of whether we want to have government controlled software design or not. Legally mandating the kind and style of redirect function available from the device manufacturer makes these base assumptions that how things are other than that is okay or that we should implicitly codify those base assumptions in law.
Is that maybe stupid? Would it better to start from some deeper principles? Maybe mandate everyone can have whatever code they want on the device, always, with no penalties for rooting, etc…