AmbitiousProcess (they/them)

all android devices have been using file based disk encryption since several Android versions now.

All Android devices are supposed to support it, but not all do. (or at least, not all do effectively without compromising the cryptographic root of trust by not implementing proper hardware security chips)

I’ll grant it to you on the scopes, PIN changes, etc, but realistically I just don’t think anyone can justify GrapheneOS being something that should be supported on Fairphone given how absolutely desolate the phone looks with regard to any attempt at all to hardware security.

It really depends on what security level you want out of a phone

It does, but that’s exactly my point. GrapheneOS will provide you essentially no more security than any other alternative Android operating system, should it have to operate on a Fairphone with all those features not supported by a Fairphone stripped away.

Unless Fairphone adds more hardware security features that are standard on most other phones, and highly supported on Pixels, installing a heavily crippled GrapheneOS on a Fairphone would get you essentially none of the benefits of GrapheneOS in the first place.

Entirely lacking disk encryption for typical users due to not having a secure element is a pretty major flaw.

It’s also missing hardware accelerated virtualization which is necessary for much of GrapheneOS’s sandboxing, has weak security for other keys in the OS keystore, is missing hardware memory tagging which makes it much easier for apps to use overflow attacks, doesn’t have proper verified boot support once a custom alternative OS is flashed, and leaves exposed debugging APIs even when the phone is locked.

This breaks:

• Secure app spawning
• Memory corruption protection
• Integer overflow protection
• Most of Graphene’s kernel hardening
• Much of Graphene’s attack surface reduction abilities
• Hardware-based attestation and security monitoring
• Quick tile protection pre-unlock
• Debugging access prevention
• Verified Boot
• The security of your PIN against any automated attack

At that point, GrapheneOS can’t physically provide you essentially any security anymore.

It wasn’t just the update cycle either, it’s that they don’t have the physical hardware chips to support Graphene’s minimum requirements for security, which would severely weaken any benefits you actually get from GrapheneOS.

Especially given it has a similar effect to how spam emails are often deliberately bad so they just don’t waste their time with people that waste some of their time before wising up. If you only get the most easily duped people in the first place, that’s a lot less time wasted.

With LLMs, you’re explicitly targeting people that are less likely to strongly ideologically object to things (like LLMs, or genocide funding) on moral grounds, you’re getting the people that are willing to just get a quick answer without having to look into it deeply, and you’re getting the people who’s sole goal it is to flood the web with any SEO-optimized slop they can with the least effort possible.

LLMs self select for people not willing to put in cognitive effort, that are easily influenced and less careful about fact-checking things.

Two is many?

And no, if you think spellcheck and language translators are slop, then unfortunately for you, a sizeable chunk of all written content ever created on the internet has been “slop” since far before LLMs even existed.

Regardless, I think it’s clear at this point you don’t give a fuck and just want to be angry, so I’ll stop and hopefully you won’t burst a blood vessel or something.

Please calm down.

“the use of LLMs to generate or rewrite article content is prohibited”

“[LLM generated] edits to existing pages may be reverted without review, and pages where the editor is the only significant contributor may be nominated for a special proposed deletion process”

Does that seem like condoning slop to you?

Again, there are only two highly limited exceptions: spellcheck (whaddaya know, the thing who’s entire purpose is to predict likely words is good at being a thesaurus and grammar checker.), and translation (you must be skilled in both used languages, re-check every source and its claims, and read through the entire article for hallucinations while crosschecking with the original document’s other language version)

This is why I stated earlier that you seem to just want something to be mad about. Wikipedia is just only allowing LLMs to be used for cases they already allowed with non-LLM tools prior, that LLMs are also more typically reliable in, while mandating extensive human review.

You are getting angry and calling me a dumbass because Wikipedia allows the narrow functions of something performing the exact same functions as spellcheck or Google Translate 💀

unlike you I don’t condone slop

Quote where I condoned slop. Oh you can’t? That’s so crazy, it’s almost as if I didn’t.

maybe don’t die on the hill trying to defend it in my proximity

Sending you links to the actual policies instead of just relying on your made up assertions of what the policies are is dying on a hill? You’re the one dying on a hill you made up lol

And btw my link was also a direct link to Wikimedia, the nonprofit who owns Wikipedia.

You linked to a blog post by Wikimedia that’s over a year old. I linked to current Wikipedia policies.

Which do you think is more reflective of the organization’s current stance:

• A single blog post written by two people over a year ago
• Multiple guideline and meta consensus documents cooperatively developed by hundreds of editors, and currently enacted as policy across all of Wikipedia.

(From the Meta-wiki page on the topic)

“Our use of AI will allow editors to focus more on what they want to accomplish, not how to technically achieve it.”

“we cannot significantly support editors with AI for both content generation and content integrity at the same time. We made a decision to first prioritize using AI to support editors to assure content integrity.”

“generative AI in particular offers a promising solution for automated mentoring and guidance of newcomers”

“we should prioritize using AI to support knowledge integrity and increase the moderator’s capacity to maintain Wikipedia’s quality”

(From the most current page regarding AI use policy on Wikipedia and the LLM Writing Guide)

“the use of LLMs to generate or rewrite article content is prohibited”

This is categorically an outright refusal to allow LLMs as a method of simply generating content. The only exceptions are for language translation (with many caveats), and for minor grammar corrections.

i.e. the shit you could do with a tool like Google Translate or Grammarly/LanguageTool/Word Spellcheck/etc, just done via a different mechanism, with anything past that being outright prohibited.

Also, it is punished, via the same means any transgression on Wikipedia is. See: https://en.wikipedia.org/wiki/Wikipedia:Disruptive_editing#Persistent_LLM_use and https://en.wikipedia.org/wiki/Wikipedia:Disruptive_editing#Dealing_with_disruptive_editors

Do you even edit Wikipedia? Have you had to rewrite or remove LLM content and put warnings on people’s talk page? Have you reviewed the live changes feed to repeatedly see LLM-generated content getting removed? Or do you just want something to complain about while ignoring what actually happens on Wikipedia because you don’t seem to even have read a single meta or guidelines page on the topic?

Exactly this. They are guaranteed to get scraped regardless. Ask any question of an AI model and the first ‘source’ it cites is almost always Wikipedia. (assuming it’s not like a super technical in-depth question)

Might as well save on server overhead costs and get paid while doing it instead of trying (and failing) to block every possible scraper.

People choosing to post slop edits on Wikipedia means Wikipedia itself is supportive of AI?

Besides, I’ve regularly seen AI generated content get labeled as LLM-generated (and marked for cleanup/removal), removed AI content myself, and seen editors revert obvious AI slop content within a minute of the edit being made. AI is absolutely not something Wikipedia or most of its editors support.

Destruction of books should not be a legal requirement, books scanned without destruction could go to a public library without violating laws.

Yeah, I wish this was the case. I don’t think that under Anthropic’s prior ruling at least, that it would be allowed though, since after donating to the library (assuming they non-destructively scanned the books) they would then have to discard the entire set of digital copies they made, as that would mean they copied the work, and thus didn’t engage in a legal “transformation” of that work that maintains only one copy, and thus there would be no reason for them to do it in the first place.

I wish our copyright law could at least consider a donation to a library as equivalent to destruction of a work so companies could keep a scan but not have it be considered a copy. That way at least we’d be getting the giant influx of books going to libraries from AI training bs.

I have no clue if that’s true, but that sounds pretty far fetched imo.

As already mentioned, the books being purchased are not like, the only sole remaining copies of the work. Many of the authors are probably still alive, and there’s likely thousands or millions of each book distributed all over the country of origin, if not the world, in the hands of individuals, bookshops, libraries, archives, etc.

A judge said it was fair use when they purchased and solely used the books for AI training without redistributing them or a copy of them afterwards, so they’re just doing that in order to not get sued again. Not much more to it.

Not a fan of the fact they’re doing this (obviously), given that I despise the waste and societal consequences AI companies are bringing upon us, but before anyone assumes they’re just doing this for the sake of being evil or something like that:

Destructive Scanning is the cheapest and easiest to automate methods of scanning books. This is fairly common for any large-scale digitizing projects that aren’t dealing with old, limited-in-supply books. (in this case, it’s just books before 2022, which still likely have many copies in circulation)

They’re also doing this to comply with court requirements for fair use. According to Ars Technica, a judge ruled that the destructive scanning qualified as fair use, only because Anthropic had spent the money to buy the books outright, destroyed them afterwards, and not distributed digital copies after.

The alternative is either Anthropic pirating the books (authors get paid $0), or buying the books, then distributing them back on the market through resellers (authors get paid the cost of the book, but then someone else buys the same copy secondhand and the author misses out on what would have otherwise been a fresh sale)

Yep. If you want to record everything going on from your POV, strap a GoPro to your head and see if people’s moods change about it. If they clearly don’t want that, they don’t want the “convenient” hidden cameras either. Simple as that.

link that doesn’t appear to be dead: https://www.news5cleveland.com/news/local-news/we-follow-through/flock-cameras-are-still-on-in-cleveland-even-though-the-contract-expired

It’s not really a “you need to bypass TCP” and more of a “TCP traffic could be censored”… just like UDP, DNS, or really any other kind of networked traffic.

Reticulum isn’t necessarily immune to this, it just supports a variety of protocols as a mesh network, so TCP isn’t something who’s failure would make the network impossible to use (but good luck accessing any traditional website without TCP).

For example, you might be able to communicate from your Android phone running a Reticulum-compatible app to a separate nearby device over Bluetooth, then that device broadcasts a signal over LoRa, which hits someone else’s LoRa-compatible radio, which then connects over a USB-C cable to their laptop, which is plugged into their router, which can then send the traffic over TCP, where it’s picked up by someone elsewhere using the internet. If TCP traffic is blocked, say, by their local government, maybe their LoRa radio just broadcasts to another LoRa radio, and another, and another, etc, until enough of them chained together is able to reach the recipient. Hence, TCP wouldn’t strictly be required, thus preventing censorship of Reticulum through blocking TCP connections. (though this would still reduce how many ways you could theoretically get to people, as if that person ONLY has access to TCP as the start of their connection to the mesh, they’d be cut off)

Of course, the government could also try jamming radio signals, then making LoRa useless, but if they do that and don’t block TCP traffic, then you still have options.

Unfortunately, I wouldn’t call Reticulum an internet replacement, nor do I think it could ever be without still relying on the kind of large-scale, high-throughput infrastructure we have for the internet today. It just doesn’t have enough bandwidth, and it’s difficult to run anything requiring low latency if every connection requires hopping through a thousand peers to get to someone on the other side of the planet who, say, wants to play the same online game as you.

Make sure to sign up via a creator’s link! (the ones they’ll put in the sponsored section of a video where they are “sponsored” by Nebula as one of Nebula’s creators)

Gets you a pretty good discount and drops it to about 30 bucks a year.

True, but that also depends on the circumstance.

Again, a lot of people just use LLMs now as their primary search engine. Google is an afterthought, ChatGPT is their source of choice. If they ask a simple question with legal or medical implications, with tons of sources, that the LLM answers with identical accuracy to those other publications, should they be sued?

I think it would be a lot better to allow people to sue if it provides false advice that ends up causing some material harm, because at the end of the day, a lot of stuff can be considered “medical.”

Maybe a trans person asks what gender affirming care is. Is that medical? I’d say it is. Should that not get discussed through an LLM if a person wants to ask it?

I’m not saying I wholeheartedly oppose this idea of banning them from giving this type of advice, but I do think there are a lot of concerns around just how many people this would actually benefit vs just cutting people off from information they might not bother to look up elsewhere, or worse, just go to less reputable, more fringe sites with less safeguards and less accountability instead.

I’m not sure I totally agree with this, even as much as I want AI companies to be held accountable for things like that.

The reason so many people turn to LLMs for legal/medical advice is because those are both incredibly unaffordable, complex, hard to parse fields.

If I ask an LLM what x symptom, y symptom, and z symptom could mean, and it cites multiple reputable sources to tell me it’s probably the flu and tells me to mask up for a bit, that’s probably gonna be better than that person being told “I’m sorry, I can’t answer that”

At the same time, I might provide an LLM with all those symptoms, and it might hallucinate an answer and tell me I have cancer, or tell me to inject bleach to cure myself.

I feel like I’d much rather see a bill that focuses more on how the LLMs come to their conclusions, rather than just a blanket ban.

Like for example, if an LLM cites multiple medical journals, government health websites, etc, and provides the same information they had up, but it turns out to be wrong later because those institutions were wrong, would it be justified to sue the LLM company for someone else’s accidental misinformation?

But if an LLM pulls from those sources, gets most of it right, but comes to a faulty conclusion, then should a private right of action exist?

I’m not really sure myself to be honest. A lot of people rely on LLMs for their information now, so just blanket banning them from displaying certain information, for a lot of people, is just gonna be “you can’t know”, and they’re not gonna bother with regular searches anymore. To them, the chatbot IS the search engine now.